7b246a7) · @caph42/pi-jev-router read and ohmyjev re-read at 0.14.0 (fbc4ea4, lockdown mode) · ohmyjev re-read at 0.13.2 (d307c39) and auto-mode 3.1.6 noted · acmelabs-15/jev-gate read, auto-mode 3.1.5 compared and ohmyjev 0.13.2 noted · ohmyjev's pi and omp gates rechecked, auto-mode and jcode read · 3 settled and 2 read , 05:17–05:40 UTC · 4 read 10 Oct · first read 6 Oct · rechecked 7 Oct · jev-approvals and jev-curator cells corrected or filled · hermes-jev-mana no-key cell read at its pin · @openclaw/typesafe 2026.10.4 release noted · failure terms aligned with the failure map and @openclaw/typesafe pre-release noted · @openclaw/typesafe 2026.10.5 release noted · @openclaw/typesafe 2026.11.1 noted, not re-read · auto-mode 4.0.0 noted, not re-read How to use Jev with Codex, OpenCode, Cursor and Hermes Agent (TypeSafe AI)
Does your coding agent's tool call or model choice still go ahead when Jev fails? 15 Jev integrations for Codex, OpenCode, Cursor, Hermes, OpenClaw, pi and omp, in code. This page answers one question for each Jev plugin, hook and MCP server for coding agents: when the Jev call fails, does the agent's tool call still run? When the pool checked search suggestions on , Google suggested "how to use jev with codex" and "how to use jev with opencode", and "jev hermes" appeared on Google, Bing and DuckDuckGo. Rechecked on : "jev codex" completes to "jev codex router" on Google, Bing and DuckDuckGo, and "jev hermes" to "jev hermes agent" or "hermes agent jev"; no new host phrase appeared on two endpoints (R12-S11). A suggestion shows that people search for a phrase, not how many do.
Checked at: ohmyjev 0.14.0 (8 Nov 2026; fbc4ea4, pi and omp adapters at their gate lines; earlier d307c39, 0.13.2, 2 Nov 2026, and 78bb82c, 21 Oct 2026); @caph42/pi-jev-router 2.0.0 75ad5b7 with pi 1.5.0 f9a2636 (read 8 Nov 2026); auto-mode 4.0.0 (9 Nov 2026; 7b246a7; earlier c0b9378, 3.0.0, 21 Oct 2026); jcode merge f5c4922 (21 Oct 2026); acmelabs-15/jev-gate b3ae111 (read 30 Oct 2026); the other rows at the pins listed in Sources (6–14 Oct 2026). Later releases noted without a re-read are in changes (auto-mode 4.0.0, noted 8 Nov 2026, was read on 9 Nov 2026 and auto-mode 3.1.6, noted 2 Nov 2026, has cited files identical to 4.0.0's; latest still not re-read: @openclaw/typesafe 2026.10.9, noted 30 Oct 2026, and 2026.10.5, noted 19 Oct 2026). Checked 2 Nov 2026: ohmyjev: re-read at d307c39 (ohmyjev 0.13.2), 2 Nov 2026: the four failure terms are unchanged in lockdown mode; the bypass cell changed in content (default lockdown-off skips every gate; auto-approve path removed; GitHub reminder added). auto-mode 3.1.6 was released on 31 Oct 2026, 09:46 UTC (npm); contents not read, not re-read (R26-S20 to R26-S24, R26-S36). Checked 30 Oct 2026, 05:20 UTC: auto-mode 3.1.5 (29 Oct 2026): no change found in its Jev call or failure handling in the files compared with 3.0.0 (a targeted comparison of the Jev path, not a full re-read; R24-S37). ohmyjev 0.13.2 (29 Oct 2026) removes the auto-approve thresholds from its policy file (R24-S38); the failure-map row was re-read at 0.13.2 on 2 Nov 2026. @openclaw/typesafe 2026.10.9 (30 Oct 2026): newer version exists, not re-read (R24-S35). Read at 4.0.0 (7b246a7, 9 Nov 2026): failure path unchanged (cited files identical to 3.1.6, 382b7a9); bypass now lists 7 skip stages, adding a reply-channel allow (src/classify-action.ts L234–242, L470–488; present since 3.1.x); still not settled because the mod host's handling of the tool.check answer was not read. No count changed (npm registry: 4.0.0 published 8 Nov 2026, 01:03 UTC; R28-S47, R29-S200 to R29-S207, R29-S17). Re-read 8 Nov 2026 at 0.14.0 (fbc4ea4): failure cells unchanged; bypass cell adds the shadow setting and the in-repo move skip. @openclaw/typesafe 2026.11.1 (5 Nov 2026) is newer than the version this page describes; not re-read. Unchanged on 7 Nov 2026: auto-mode 3.1.6. No cell or count changed; the failure map still counts n = 80 (npm and PyPI registry records, 7 Nov 2026, 15:29–15:36 UTC; R27-S25).
Of 29 coding-agent integrations read in source, 15 act on Jev's answer, 10 do not and 4 are unsettled. pi-jev-router (pi) picks each request's model and uses the standard model when Jev fails.
ohmyjev 0.14.0 (read 8 Nov 2026) keeps its gates off by default and adds a shadow setting that logs what lockdown would block without blocking it. Documented source at fbc4ea4; R28-S20 to R28-S23
Picked from an earlier lead list, the Hermes Agent catalog, three GitHub searches, daily scans and, on 10 Oct, 3 pinned reads chosen by host priority from 35 scan leads; on 11 Oct, 2 of the README-only candidate gates read at pins; on 19–21 Oct, ohmyjev's pi and omp adapters, auto-mode and jcode; on 30 Oct, acmelabs-15/jev-gate; on 8 Nov, @caph42/pi-jev-router; not a survey. The 29 are coding-agent integrations. Of the 29, 1 is vendor-published (OpenClaw's plugin), 28 are community projects and 0 come from TypeSafe. On a Jev error the 15 that act are 3 fail-closed, 10 fallback and 2 fail-open (failure-map terms, 21 Oct 2026; counts updated 8 Nov 2026). These are counted on this page; the failure map's 80 is unchanged. The pool installed and ran nothing: no agent, hook, plugin, launcher, proxy or MCP server was started and no Jev call was made. npm tarballs were downloaded and read, not executed; jcode's source was read at its pinned commit and not built or run.
- 3 fail-closed: jev-kit, bloudhood's Codex approval hook and jev-curator (guard mode) hold the tool call.
- 10 fallback, each to a named path: opencode-tool-gate to OpenCode's permission rules; osuki to its default tier; jev-gateway to the agent's own model, which picks the tool; jev-approvals to Hermes's main agent model on a timeout, connection error, 429 or 402 when a main provider is set (that model can approve the command; other errors go to a person); rh-guard to its lexical detectors; and 5 model-choice or routing rows: jev-effort-router (configured model), jev-triage (main model, OpenClaw), the jev-for-all launcher (default tier: standard, claude sonnet, medium effort), hermes-jev-mana (original Hermes branch) and pi-jev-router (standard model, pi; under Read in source; not on the failure map yet).
- 2 fail-open: jev-judge (enforce mode) and ohmyjev's pi and omp adapters let the call run.
Does the tool call still run on a Jev error? Yes in 5 of 15 (2 fail-open, 3 fallback: OpenCode's permission rules, the default tier, the agent's own model), conditionally in 2 (both fallback), no in 3 (fail-closed). The other 5 are model-choice or routing rows (fallback): the turn runs on another model.
3 fail-closed + 10 fallback + 2 fail-open = 15 counted; 5 yes + 2 conditional + 3 no + 5 routing = 15. 10 not counted (8 return Jev's answer to the agent + 2 give advisory hints) and 4 not settled (Hoppin-Sorter/jev-router, auto-mode, jcode, acmelabs-15/jev-gate): 15 + 10 + 4 = 29 (counted on this page; the failure map's 80 is unchanged). To compare your own design with the 80 implementations on the failure map, use the failure self-check.
3 of the 14 rows in the table below (jev-approvals, jev-curator, hermes-jev-mana) do nothing until you switch them on. jev-approvals is off after install: Hermes core's default approval mode is manual, so you must set approvals.mode: smart. jev-curator runs in observe by default: Jev does not block anything until you change the mode to guard or apply. hermes-jev-mana is inert after the documented install: every Jev call is refused because admit_unlabeled defaults to false; it acts only if the operator sets admit_unlabeled: true. Two more start in a softer mode: jev-judge only logs (shadow) until you switch on enforce, and osuki uses Jev only if you pick TypeSafe as its decision provider. jev-effort-router's routing is on in the plugin's own config, but the plugin is off until it is enabled in Hermes; with Hermes defaults (model "") it is inert, because it acts only when the provider is Ollama:Cloud (R13-S217). jev-gateway routes by default once a launcher points the agent at it. jev-triage is on at startup and acts where Jev is the agent's decision model. The jev-for-all launcher acts only when you run jev-for-all start. ohmyjev's pi and omp adapters deny the calls they flag and never add a prompt once lockdown is on and shadow is off; both settings are off by default (source read 8 Nov 2026); the no-prompt behaviour is from its README (Reported). Changed 11 Oct 2026: was "2 of the 10"; 21 Oct 2026: of the 14
Does the tool call still run when Jev fails? The 15 integrations that act on Jev's answer
One row per implementation that acts on Jev's answer itself: 14 rows in this table, and the 15th, @caph42/pi-jev-router (pi, counted 8 Nov 2026), under Read in source; not on the failure map yet. By host: Codex 4 (jev-kit, bloudhood, rh-guard, jev-gateway), Cursor 1 (rh-guard), OpenCode 4 (opencode-tool-gate, osuki, jev-gateway, jev-for-all launcher), Hermes Agent 5 (jev-approvals, jev-curator, jev-judge, jev-effort-router, hermes-jev-mana), OpenClaw 1 (jev-triage), pi 2 (ohmyjev, pi-jev-router) and omp 1 (ohmyjev). rh-guard covers Codex and Cursor, and jev-gateway covers Codex and OpenCode (and also Claude Code, Kilo, Gemini and Devin, named in its row only); the jev-for-all launcher also starts Claude Code (named in its row only). ohmyjev's pi and omp adapters are one row because they share one gate path. So the host numbers add up to 18 for 15 counted rows. All 7 hosts have a counted row; OpenClaw has one since 11 Oct (jev-triage), pi and omp since 21 Oct (ohmyjev's adapters). ohmyjev's Claude Code module is not a row here: it is on Claude Code and MCP guards and is already one of the 80 on the failure map, whose ohmyjev row now names pi and omp, so the map count does not change. jev-effort-router, the jev-for-all launcher and hermes-jev-mana's routing points decide a model choice, not a tool call; jev-triage decides whether an inbound message reaches the main model. The tool calls of those turns still run. Columns (a) to (e) are the five failure cases: (a) no API key, (b) the Jev call times out, (c) Jev returns an HTTP error, (d) the answer is malformed, (e) Jev is unsure (below the integration's threshold). Cells use the pool's shared failure vocabulary. Unread cells say not recorded. Documented source at the pins in Sources. 6 rows read 6 Oct 2026, 05:14–05:29 UTC; jev-approvals and jev-curator settled and the rh-guard, opencode-tool-gate and osuki cells rechecked 7 Oct 2026, 05:17–05:27 UTC. On 8 Oct 2026, 05:14–05:16 UTC, the jev-approvals timeout and HTTP-error cells were corrected, jev-curator's three open cells filled and rh-guard's low-confidence text clarified (changes); Hermes core auxiliary_client.py L7625–7675 spot-checked against the raw file at the pin. jev-effort-router and jev-gateway read at the pinned commit on 10 Oct 2026, 05:12–05:16 UTC; jev-effort-router's below-threshold cell spot-checked against the raw file at the pin (05:19:56 UTC). jev-triage and hermes-jev-mana settled and the jev-for-all launcher read at the pinned commit on 11 Oct 2026, 05:17–05:27 UTC; OpenClaw choose-route.ts L618–642 (R13-S204) and the hermes-jev-mana patch L88–93, L428–429 and L1396–1397 (R13-S210) spot-checked against the raw file at the pin. Spot-checked against the raw files at the pins: bloudhood, rh-guard, jev-judge (6 Oct) and Hermes core approval.py, approval_smart.py and OpenCode agent.ts (7 Oct). jev-kit and opencode-tool-gate were read and checked line by line at the pinned commits; osuki and jev-curator were read once at the pin. ohmyjev's pi and omp adapters were read at the pinned commit on 19 Oct 2026 and their gate path re-read at the same commit on 21 Oct 2026, 05:11–05:13 UTC; the gate return, no-key and timeout lines spot-checked against the raw files at the pin (R21-S101, R21-S103, R21-S105)
| Integration (pin) | Hosts and hook point | Publisher | Mode after install | (a) No key | (b) Timeout | (c) HTTP error | (d) Malformed | (e) Low confidence | Bypass | Tool call still runs on a Jev failure? |
|---|---|---|---|---|---|---|---|---|---|---|
keiffff/jev-kit jev-agent-review3d74b05 (5 Oct 2026, 19:00 UTC); npm @jev-kit/cli 0.3.1 | Codex and Claude Code; PermissionRequest command hook (4 s hook timeout in both examples) | Community (keiffff); MIT | Enforcing once the hook is configured, approve-only: it auto-approves or defers to the host's own prompt and never denies | fail-closedHeld for the host's own prompt | fail-closedSDK default 10 s, no retries; held for the host's prompt | fail-closedHeld (defer) | fail-closedHeld (defer) | heldAllow only if policy-compliant ≥ 0.7, instruction-aligned ≥ 0.7 and high-risk ≤ 0.15 (example policy), else defer | recordedOnly permission requests reach it; secret-like input is deferred without a Jev call | NoThe host asks you (a–d) |
bloudhood/codex-jev-approval-hookb6242c9 (pushed 29 Sep 2026) | Codex; PermissionRequest hook for Bash, apply_patch and mcp__ tools; 12 s host timeout; PowerShell | Community (bloudhood); MIT | Enforcing for allow and deny; otherwise Codex's own approval | fail-closedHeld for Codex's own approval | fail-closedDefault 5 s (1–5 allowed), 9 s budget; held | fail-closedHeld | fail-closedHeld | heldAllow needs ≥ 0.98 with deny ≤ 0.01; deny needs ≥ 0.80 (author's constants) | recordedOnly Bash, apply_patch and mcp__; stops deciding after 3 denials in a row or 20 per session | NoCodex asks you (a–d) |
anpicasso/hermes-jev-approvals jev-approvals v0.3.0Plugin 28be98a (catalog commit) with Hermes Agent core v2026.9.24 = f97608f (24 Sep 2026, latest release on 7 Oct) | Hermes Agent; answers Hermes core's smart-approval question for commands Hermes flags as dangerous | Community (anpicasso); MIT | Off after install. Hermes core's default approval mode is manual; it acts only with approvals.mode: smart. Then an APPROVE runs the command without a person (core tools/approval.py L774–777) | fail-closedThe plugin raises; Hermes core escalates to a person (unless auxiliary.approval.fallback_chain is set) | fallback-main agent modelOn a timeout or connection error when a main provider is set (an explicit fallback_chain is tried first); that model can APPROVE (with no main model configured, it asks you). Plugin deadline 25 s; core may retry twice more Corrected 8 Oct 2026; was fail-closed. "(with no main model configured, it asks you)" added 11 Oct 2026 (R13-S217) | fallback-main agent modelOn a 429 or 402 when a main provider is set (with no main model configured, it asks you); held (escalate) on other HTTP errors (401, 403, 404, 422, 5xx, 529; approval_smart.py L126–131) Corrected 8 Oct 2026; was fail-closed | fail-closedAn empty or unknown answer escalates (L115–125) | heldAn APPROVE below confidence 0.55 becomes ESCALATE (plugin jev_policy.py L13, L54–59) | recordedMode off or yolo; permanent allowlist; commands with no dangerous or tirith warning never reach it; non-interactive contexts skip the smart step | ConditionalOn a timeout, connection error, 429 or 402 with a main provider set, the main agent model answers and an APPROVE runs the command without a person (b–c). No key, a malformed answer and other HTTP errors go to a person; with no answer the gateway or CLI denies (a, c, d) Changed 8 Oct 2026; was "No" |
anpicasso/hermes-jev-curator jev-curator v0.1.04e8626c (catalog commit; its own repository) | Hermes Agent; pre_tool_call guard on skill_manage calls from Hermes's background review, plus a skill lifecycle hook that starts a dry-run scan | Community (anpicasso); MIT | Observe by default: Jev judges skills, but nothing is blocked until you set mode guard or apply. In those modes, plans built from Jev's judgments decide which background skill deletes or patches may go ahead, with no person involved. The deletes come from Hermes core's background review; the plugin does not start them | fail-closed"No credential is configured" is raised (transport.py L118–122); that pair gets no judgment, so a background delete or patch it would cover is blocked (guard.py L144–160) Filled 8 Oct 2026 | fail-closed25 s; that pair gets no judgment, so a background delete or patch it would cover is blocked | fail-closedNo judgment, no plan: blocked | fail-closedNo answers object or invalid answers raise (transport.py L128–131, L146–191): no judgment, no plan: blocked Filled 8 Oct 2026 | heldBelow MIN_CONFIDENCE 0.85 the edge is refused as "low-confidence" and not authorised, so the guard blocks (graph.py L21–24, L165–176) Filled 8 Oct 2026 | recordedModes off and observe leave the guard inert; foreground skill operations are never gated; creates and unknown actions pass; other tools are ignored; an exception in the outer hook lets the call through | NoIn guard mode, background deletes and patches not covered by a plan are blocked (a–d), and low-confidence judgments do not authorise them (e); other skill operations are unaffected |
24601/rh-guardc2e682e (pushed 28 Sep 2026) | Codex (PreToolUse) and Cursor (preToolUse, beforeShellExecution) | Community (24601); MIT | Shadow, review or enforce (RH_GUARD_MODE). Default enforce: unset, empty or unknown values give enforce (kinds.ts L176–189), so a default install can block | fallback-lexical detectors | fallback-lexical detectors1,200 ms | fallback-lexical detectors | fallback-lexical detectorsUnparseable answers dropped | acts-anywaySteer message only when Jev's primary_kind confidence is below 0.5 on a mutating step (score.ts L110–116); the call still runs unless deny-kind labels reach a block threshold (L98–106). REVIEW_THRESHOLD 0.45 (kinds.ts L36) is a risk-label floor for steering, not a confidence threshold Text clarified 8 Oct 2026; term unchanged | recordedCodex matcher omits MCP tools; a structural deny skips Jev; shadow mode (opt-in) returns "ok" | ConditionalIn the default enforce mode the call runs on a Jev error, with a steer message on mutating steps, unless the structural or lexical detectors alone reach a block threshold (0.70–0.88 by kind); then it is denied (score.ts L98–109, L174–181) |
justmytwospence/opencode-tool-gate 0.1.0bdf1ef8 (6 Oct 2026, 02:59 UTC) | OpenCode; tool.execute.before plugin hook | Community (justmytwospence); MIT | Enforcing by default (enabled: true) | fallback-opencode permission rulesThe gate allows the call; OpenCode's own rules apply | fallback-opencode permission rules3,000 ms default | fallback-opencode permission rules | fallback-opencode permission rulesMissing fields lead to allow | acts-anywayHeld only above set cuts (for example irreversible ≥ 0.9, exfiltration ≥ 0.8); user-configurable | recordedRead-only tools and allowTools skip Jev; an always-held list (sudo, force-push main, curl | sh and others) is held without Jev | YesOpenCode's default rules allow every tool ("*": "allow"); they ask only for paths outside the project, repeated identical calls and .env reads. Your own permission config can change thisOpenCode v1.18.35 = 53d1eab, packages/opencode/src/agent/agent.ts L119–136 |
@osuki-dev/opencode-osuki-agentPinned at 0.2.7 (npm, 4 Oct 2026); tag v0.2.7 = c96ce03; tarball and tag not matched file by file. Checked at 0.2.8 (npm, 5 Oct 2026, 11:40 UTC; tag v0.2.8 = 0ae9e88; 7 Oct 2026): failure path unchanged; the hook lines moved by 9. The 0.2.8 tarball was not matched to its tag file by file either | OpenCode; reroutes the subagent tool and prunes the tool list | Community (osuki-dev); MIT | Enforcing (router). Jev is opt-in: the default decision provider is opencode.ai/zen; TypeSafe needs TYPESAFE_API_KEY | fallback-default tier"standard" | fallback-default tier10,000 ms, then a cooldown | fallback-default tier60 s cooldown skips Jev | fallback-default tierTool pruning falls back to the full tool list | fallback-default tierBelow 0.75 (routing) or 0.8 (tools); plugin options | recordedFixed roles (plan, review, explore) skip Jev; a previous decision is reused | YesThe subagent runs on the default tier with all tools (a–d) |
DoGMaTiiC/hermes-jev jev-judge5dddbea (25 Sep 2026; catalog commit = HEAD). The same repository ships jev-skill-router, an advisory skill hint, listed here and not counted separately | Hermes Agent; pre_tool_call hook plus a jev_ask tool | Community (DoGMaTiiC); MIT; Hermes catalog "Community" | Shadow (log-only) by default. The cells to the right describe enforce mode, where it asks you to confirm terminal, write_file and patch calls it flags | fail-open | fail-open3.0 s default, capped at 10 s | fail-open | fail-open | acts-anyway"Clear" unless destructive ≥ 0.90, exfiltration ≥ 0.70 or impact ≥ 2.5 (user-configurable) | recordedUnlisted tools; 300 s cache; circuit breaker after three 429/529 responses (fail-open); any hook exception fails open | YesIn enforce mode (a–d); in the default shadow mode it never stops a call |
AlphaPerseii3000/jev-effort-router v0.2.2562dc2a (HEAD on 8 and 10 Oct; the Hermes catalog at main dce1e9b pins it). Added 10 Oct 2026 | Hermes Agent; llm_request middleware (__init__.py L68). Rewrites model and reasoning_effort in the outgoing request once per user turn; Hermes sends the rewritten request without a person (host read at f97608f). Route: OpenRouter, model typesafe/jev-1.13 | Community (AlphaPerseii3000) | Routing on in the plugin's own config (plugin.yaml L33–35), but off until enabled in Hermes: Hermes plugins are opt-in, and hermes plugins install asks "Enable now? [y/N]" only on a terminal (plugins_cmd.py L986–994, R13-S220). Inert until OPENROUTER_API_KEY is set. With Hermes defaults (model "") it is inert, because it acts only when the provider is Ollama:Cloud (R13-S217) Wording corrected 11 Oct 2026; was "Routing on in the plugin's defaults … whether Hermes needs a separate enable step: not recorded" | fallback-configured modelThe request is left unchanged (client.py L178–179; router.py L250–253) | fallback-configured model2.0 s (config.py L31) | fallback-configured modelAny non-2xx (client.py L204–208) | fallback-configured modelFor a non-JSON body or no answers. A partial answer with no model answer gives deepseek-v4.1-flash (fallback-default model) | fallback-default modelBelow 0.5 or an unknown choice, the code rewrites the request to deepseek-v4.1-flash at medium effort (client.py L237–252, L279–285; router.py L182–184), unless the provider catalog lacks that model. This differs from the plugin's own docstring and catalog text | recordedenabled: false, another provider or API mode, an off-grid configured model, no user message, settings unavailable, any exception | Runs on the configured model(a–d); on deepseek-v4.1-flash at medium effort (e). A model-choice row: the turn's tool calls run in every case |
jev-gateway 0.5.1npm 0.5.1; repository named in npm metadata (gitHead 37ff982); code read from the npm 0.5.1 tarball, read in memory, not executed. Added 10 Oct 2026 | Codex and OpenCode (also Claude Code, Kilo, Gemini, Devin); a local proxy between the agent and its LLM. It forces tool_choice to Jev's pick (decide.js L161) or builds the tool call itself (L150–157); with the default JEV_ON_NONE=force_none Jev can force "no tool" | Community (npm package) | Routing on by default (config.js L44); direct calls on (L43) | fail-closedThe gateway does not start (jev.js L65–66; launcher.mjs L112–119) | fallback-agent's own model4,000 ms plus 1 retry on 408/429/5xx (about 8.1 s at worst); the request passes through (decide.js L102–105) | fallback-agent's own modelPassthrough (decide.js L102–105) | fallback-agent's own modelPassthrough (decide.js L106–110) | fallback-agent's own modelBelow 0.7, or when the two Jev answers disagree, the request passes through and the agent's model picks the tool (L122–127) | recordedheader x-jev-gateway: off, routing off, the Codex auto-reviewer model, no tools or an already-decided tool choice, hosted or namespaced tools; an upstream 400/422 replays the original request | Yes(b–e): the agent's own model picks the tool and the call runs. No on (a): the gateway does not start. The host's own approval path after the proxy was not read |
RanyAlbegWein/jev-triage 1.0.36360e21; host OpenClaw aa6008a (provenance commit of @openclaw/typesafe 2026.9.8). Settled and counted 11 Oct 2026 (not settled on 10 Oct) | OpenClaw; before_dispatch hook that triages inbound messages. At 0.9 or above, "thanks" gets a canned reply and "spam" ends the turn silently (index.js L45–47). With handled set, the host sends the plugin's text as the final reply and returns before the agent dispatch, so the main model is not called and no person is involved (choose-route.ts L618–642, before L657 and L663–692). "question" and "incident" only add an advisory note | Community (RanyAlbegWein) | On at startup (openclaw.plugin.json "activation": onStartup). Acts only where Jev is the agent's decision model; the README install sets agents.defaults.decisionModel and requires hooks.allowConversationAccess true Reported | fallback-main modelAny non-ok status returns undefined and the message goes to the main model (index.js L41) | fallback-main model5,000 ms (L39); an abort or throw is caught by the host, where before_dispatch is fail-open (hooks.ts L829–840; hook-runner-global.ts L22–30) | fallback-main model(L41) | fallback-main modelA throw at L42–44 is caught by the host Settled 11 Oct 2026; was not recorded | fallback-main modelBelow 0.9 the message goes to the main model untouched (L45) Term changed 11 Oct 2026; was acts-anyway, same outcome | recordedempty text or a leading / (L34); host inbound handlers off or the session runtime restricted (choose-route.ts L558–562) | Goes to the main model(a–e): the message is handled as if the plugin were absent. A routing row: no tool call is gated |
jev-for-all 0.7.1, start launchernpm 0.7.1 (gitHead 74f62e7); code read from the npm tarball in memory, not executed. Added 11 Oct 2026 (was README only) | Claude Code and OpenCode; jev-for-all start "<prompt>" asks Jev for tier and effort, then launches claude or opencode with that model, with no person involved (bin/pick.ts L61–72; src/models.ts L20–75) | Community (npm package) | Acts only when you run start. Counted on the start launcher; the OpenCode plugin narrows the tool list and is not settled (OpenCode's handling of removed tools not read); the Claude Code hook was not read | fallback-default modelStandard tier (claude sonnet), medium effort (pick.ts L61–63) | fallback-default model3,000 ms (pick.ts L62); standard tier, medium effort (models.ts L40–41, L69–72) | fallback-default modelStandard tier, medium effort (same lines) | fallback-default modelStandard tier (L32–33) | fallback-default modelStandard tier unless light ≥ 0.6 or heavy ≥ 0.5; medium effort unless the top effort ≥ 0.5 (L45–56) | recordedempty prompt; the user's models.json overrides | Runs on the default tier(a–e): standard tier (claude sonnet), medium effort. A model-choice row: the session's tool calls run in every case |
Novacon/ohmyjev, pi and omp adapters in lockdown mode without shadow (both off by default)fbc4ea4 (0.14.0; read at the gate lines 8 Nov 2026); before that d307c39 (0.13.2; read at the gate lines 2 Nov 2026). Earlier 78bb82c (HEAD since 15 Oct 2026; unchanged on 21 Oct): read at the pinned commit 19 Oct 2026; gate path re-read at the same commit 21 Oct 2026. Counted 21 Oct 2026 (lead triage line 51 until then). Its Claude Code module is on Claude Code and MCP guards. Noted 30 Oct 2026: ohmyjev 0.13.2 (29 Oct 2026) removes the auto-approve thresholds from its policy file (R24-S38). Change, 2 Nov 2026: re-read at d307c39 (ohmyjev 0.13.2), 2 Nov 2026: the four failure terms are unchanged in lockdown mode; the bypass cell changed in content (default lockdown-off skips every gate; auto-approve path removed; GitHub reminder added). The gates run only with lockdown on; lockdown defaults to off (policy.ts L12; .claude-plugin/plugin.json L29–34). No count changed (R26-S20 to R26-S24). Change, 8 Nov 2026: re-read at fbc4ea4 (ohmyjev 0.14.0): the four failure terms are unchanged (the cells keep the line numbers read at d307c39; at fbc4ea4 only line numbers move); the bypass cell adds the shadow setting, off by default, which runs the lockdown checks but only logs what they would block, and the in-repo mkdir, touch, cp and mv skip. No count changed (R28-S20 to R28-S23) | pi (tool_call, adapters/pi.ts from L498) and omp (tool_call, adapters/omp.ts from L524); both use adapters/shared.ts and the same hooks/policy.ts; gates need lockdown (pi.ts L508, L522, L538; omp.ts L535, L563, L590, at fbc4ea4) | Community (Novacon) | Enforcing in lockdown mode without shadow (both off by default). With lockdown on and shadow off, denies the Bash, write/edit and exfiltration calls it flags; never adds a prompt. "If Jev is unsure, slow, or down, the call goes through like normal" (README L110 at 78bb82c) Reported | fail-openaskJev throws (shared.ts L28); the gate returns without blocking (pi.ts L512, L529, L545; omp.ts L552, L580, L595); both log "no Jev key: gates and screens let every call through" (pi.ts L449; omp.ts L413) | fail-open1,500 ms (policy.ts L84; shared.ts L37); same return | fail-openThe error throws and decide returns nothing (shared.ts L106); same return | fail-openparseReply throws (jev.ts L57–66); same return | acts-anywayBash denied only at irreversible ≥ 0.6, irreversible with destructive intent ≥ 0.7 or exfiltration ≥ 0.7; writes only at secret ≥ 0.7 (secrets file ≥ 0.8) (policy.ts L390–403, L405–428; thresholds L57–58, L65) | recordedLockdown off, the default, skips every gate before any Jev call (pi.ts L508, L522, L538; omp.ts L535, L563, L590 at fbc4ea4); the shadow setting (off by default) runs the gates and Jev's checks but only logs what they would block, so every call runs (pi.ts L504, L602–606; omp.ts L531, L678–682; shared.ts L173–182 at fbc4ea4); in-repo mkdir, touch, cp and mv skip Jev unless policies are set (policy.ts L370, L414 at fbc4ea4); other line numbers in this cell as read at d307c39: with lockdown off, a git push or GitHub write is denied once by a code reminder and runs when repeated (pi.ts L503; omp.ts L529; no Jev call); off switch (pi.ts L499; omp.ts L525); read-only commands skip Jev (pi.ts L509; omp.ts L534); writes outside allowed paths denied by code (pi.ts L525; omp.ts L571); each gate can be switched off; other tools never sent to Jev (pi.ts L546–547; omp.ts L596 at 78bb82c; not re-read at d307c39); its own exceptions pass the call | YesThe gate returns without blocking, and the host runs the tool unless a tool_call handler blocks: pi and oh-my-pi document that such a handler "may block" (pi docs/extensions.md L218 at 16f5f0a; oh-my-pi docs/extensions.md L548 at d5cd750) Documented; that it "doesn't ask" is from ohmyjev's README (L176) Reported. pi's and omp's own code after the return was not read |
abehrman/hermes-jev-mana3daf639, with a core patch (patches/jev-decisions.patch, 2,079 lines, read in full) targeting Hermes e97923c3. Settled and counted 11 Oct 2026 (not settled on 10 Oct) | Hermes Agent; decision points patched into the host. When admitted, Jev's answer picks the model lane and reasoning effort, can make the assistant's final answer without a provider call, can run an executor (claude -p, codex exec, opencode run) without a person, and builds tool calls from a chosen plan (patch L1116–1201, L1782–1836). Its approval point can only escalate to a person | Community (abehrman) | Inert after the documented install: every Jev call is refused because admit_unlabeled defaults to false; acts only if the operator sets admit_unlabeled: true. The installer turns jev.decisions on (jev_mana_install.py L334–337); each refused point records local_privacy_skip and runs the original Hermes code (patch L93, L428–429) | fallback-original Hermes branchJudgement: the plugin shells out to the jev CLI; every CLI failure raises JevError, which the published error cell maps to the original branch. The key is read inside the CLI, which was not read (plugin/jev-mana/jevkit/client.py L194–226; read 14 Oct 2026; R16-S128) | fallback-original Hermes branchOn the configured model. 1.0 s per call / 5.0 s per turn + 0.25 s grace (patch L89–90, L1396–1397); "outage" leads to fallback: original (L436–438) | fallback-original Hermes branchOn the configured model. Any exception becomes "outage" (jevkit/decisions.py L282–287) | fallback-original Hermes branchOn the configured model. "invalid" (decisions.py L284–285; patch L442–458, L1062–1064) | fallback-original Hermes branchOn the configured model. Near-tie margin 0.15 in code (patch L88; the example config comment says 0.10) gives "uncertainty" (L1069–1078) | recordeddecisions off; plugin not enabled; excluded points; admission false; explicit model or effort pins; delegated agents get no executor (L937–941); fewer than 2 options; more than 8 native steps per turn; unhealthy points (decision_hooks.py L61–71) | Runs the original Hermes branch(a–e), on the configured model; (a) is a judgement. Inert after the documented install (see Mode) |
- fail-closed or held: the tool call waits for you or the host's own prompt
- fail-open or acts-anyway: the tool call can run; "conditional" cells that can act count here
- fallback-<what>: a named substitute decides
- no-decision: nothing returned; the caller must handle it
- advisory: Jev never decides
- not recorded or not applicable
Changes on 9 Nov, 8 Nov, 2 Nov, 30 Oct, 23 Oct, 21 Oct, 18 Oct, 11 Oct, 10 Oct and 8 Oct 2026
- Change, 9 Nov 2026: auto-mode row read at 4.0.0 (
7b246a7): five cells unchanged; the bypass cell now lists 7 skip stages, adding a reply-channel allow (present since 3.1.x); still not settled; counts unchanged (15 + 10 + 4 = 29); the failure checklist still counts 80. Documented R29-S200 to R29-S207, R29-S17 - Change, 8 Nov 2026: @caph42/pi-jev-router 2.0.0 (pi, read at
75ad5b7with pi 1.5.0 atf9a2636) added under Read in source; not on the failure map yet, counted. Read in source 28 → 29; counted 14 → 15; community 27 → 28; on a Jev error fallback 9 → 10 (3 fail-closed and 2 fail-open unchanged); routing rows 4 → 5; host pi 1 → 2; not counted 10, not settled 4, vendor-published 1 and from TypeSafe 0 unchanged. Counted on this page; the failure map's 80 is unchanged. ohmyjev's pi and omp row re-read atfbc4ea4(0.14.0) in lockdown mode without shadow (both off by default); terms unchanged; the bypass cell adds the shadow setting and the in-repo mkdir, touch, cp and mv skip; no count changed. auto-mode 4.0.0 (8 Nov 2026) noted, not re-read. 19 lines added to lead triage (24 Oct–8 Nov 2026). Documented R28-S200 to R28-S210, R28-S20 to R28-S23, R28-S48 - Change, 2 Nov 2026: ohmyjev's pi and omp row and the failure-map row re-read at
d307c39(0.13.2) in lockdown mode (its default is off); terms unchanged; no count changed. The bypass cell changed in content (default lockdown-off skips every gate; GitHub reminder added). auto-mode 3.1.6 was released on 31 Oct 2026, 09:46 UTC (npm); contents not read, not re-read; the row's cells are unchanged. Documented R26-S20 to R26-S24, R26-S36 - Added 30 Oct 2026: acmelabs-15/jev-gate (Claude Code
PreToolUsehook, read atb3ae111), not settled. Read in source 27 → 28; not settled 3 → 4; community 26 → 27; counted 14, not counted 10, vendor-published 1 and from TypeSafe 0 unchanged. Not on the failure map (n 80 unchanged). 10 lines from the 21–30 Oct scans added to lead triage. Documented R24-S228 to R24-S245, R24-S43 - Rechecked 30 Oct 2026: auto-mode 3.1.5 was released on 29 Oct 2026; a comparison with 3.0.0 found no change to its Jev call or failure handling in the files compared (a targeted comparison of the Jev path, not a full re-read); it still requests
jev-1.13.0by ID. ohmyjev 0.13.2 (29 Oct 2026) removes the auto-approve thresholds from its policy file; the pi and omp row and the failure-map row were read at an earlier version and have not been re-read at 0.13.2.@openclaw/typesafe2026.10.9 (30 Oct 2026): newer version exists, not re-read. No cell changed. Documented R24-S35, R24-S37, R24-S38 - Change, 23 Oct 2026: the jev-gateway repository link now points to
vinilana/jev-gatewayat the same commit37ff982, the repository named in its npm metadata; the earlier link returned HTTP 404 on 23 Oct 2026, 05:34 UTC. No cell changed. R22-S52, R22-S53 - Added 21 Oct 2026: 1 counted row, ohmyjev's pi and omp adapters (
78bb82c; fail-open on a Jev error), and 2 not settled, auto-mode 3.0.0 and jcode (mergef5c4922). Read in source 24 → 27; counted 13 → 14; on a Jev error fail-open 1 → 2 (3 fail-closed and 9 fallback unchanged); the tool call still runs, yes 4 → 5; not counted 10 (unchanged); not settled 1 → 3. ohmyjev's Claude Code module was already one of the 80 on the failure map; that row now names pi and omp, and the map still counts 80. Lead triage line 51 (ohmyjev, README only) now points to the row; 13 lines from the 19–20 Oct scans added to lead triage. Documented R21-S100 to R21-S112, R21-S115 to R21-S138, R21-S200 to R21-S231 - Change, 18 Oct 2026: the key finding's error split "5 let the call run, 3 hold, 5 fall back" could not be rebuilt from the table's own cells. It now uses the failure map's terms for the same 13 rows: 3 fail-closed, 9 fallback, 1 fail-open; whether the tool call still runs is a separate line (yes 4, conditional 2, no 3, routing 4). The figure's error, no-key and malformed rows use the same terms. No cell outcome changed: the 13 rows were compared cell by cell with the failure map (0 outcome differences). Wording only: the 13 Bypass cells start with the map's recorded chip, and hermes-jev-mana's (b)–(e) cells read fallback-original Hermes branch, with "on the configured model" in the detail. Documented R18-S63
- Rechecked 19 Oct 2026:
@openclaw/typesafe2026.10.5 published 18 Oct 2026; not re-read; cells rest on the pinned read. R19-S50 - Rechecked 18 Oct 2026:
@openclaw/typesafe2026.10.4 is still the latest release at 05:23 UTC; a pre-release2026.10.10-beta.2is listed; not re-audited. R18-S57 - Counts changed 11 Oct 2026: read 22 → 24; counted 10 → 13 (jev-triage, jev-for-all, hermes-jev-mana); not counted 10 (unchanged); not settled 2 → 1 (Hoppin-Sorter/jev-router). Documented R13-S200 to R13-S228; read at the pin 05:17–05:27 UTC; spot-checked against the raw files: R13-S204, R13-S210, R13-S217
- On a Jev error, changed 11 Oct 2026: held 3/10 → 3/13; the call runs 5/10 → 5/13; fall back to a model or the original path 2/10 → 5/13. Below threshold: held 4/10 → 4/13, acts-anyway 3/10 → 3/13, fallback 3/10 → 6/13.
- Corrected 11 Oct 2026: jev-triage's below-threshold term acts-anyway → fallback-main model (same outcome; routing rule) and its malformed cell not recorded → fallback-main model. jev-effort-router's mode text: routing is on in the plugin's own config, the plugin is off until enabled in Hermes, and with Hermes defaults (model
"") it is inert (R13-S217, R13-S219, R13-S220); its failure cells hold as written. jev-approvals: "(with no main model configured, it asks you)" added. - README-only candidate gates 8 → 6 on 11 Oct 2026; 31 more scan lines triaged (lead triage).
- Rechecked 16 Oct 2026:
@openclaw/typesafe2026.10.4 (patch, calendar version) is the newest release on 16 Oct 2026, 05:28 UTC; not re-audited. R17-S57 - Rechecked 14 Oct 2026:
@openclaw/typesafe2026.10.3 released 13 Oct 2026, 12:36 UTC; not re-audited. R16-S64 - Change, 14 Oct 2026: hermes-jev-mana's no-key cell changed from not recorded to fallback-original Hermes branch (judgement), read at the pinned commit
3daf639. No API key for the 13 counted: hold, block or do not start 5, run 4, fall back 3 → 4, not recorded 1 → 0 (5 + 4 + 4 = 13). Same cell as on when Jev fails. R16-S128 - Rechecked 12 Oct 2026:
@openclaw/typesafe2026.10.2 released 12 Oct, 02:27 UTC; the jev-triage host branch was last read at 2026.10.1 (not re-read). R14-S53 - Rechecked 11 Oct 2026:
@openclaw/typesafe2026.10.1 released 10 Oct; the jev-triage host branch read at 2026.10.1 up to L650 has the same effect. R13-S26, R13-S208, R13-S209 - Added 10 Oct 2026: 2 counted rows, jev-effort-router (Hermes Agent,
562dc2a) and jev-gateway (npm 0.5.1). Read in source: 18 → 22; counted 8 → 10; not settled 0 → 2 (jev-triage, hermes-jev-mana); not counted 10 (unchanged). Documented R12-S200 to R12-S209, R12-S254 to R12-S257; 05:12–05:20 UTC - Counts changed 10 Oct 2026: on a Jev error, held 3/8 → 3/10, may ask the main model 1/8 → 1/10, the call runs 4/8 → 5/10 (jev-gateway), keeps its configured model 0 → 1/10 (jev-effort-router). Below threshold: held 4/10, acts-anyway 3/10, fallback 1/8 → 3/10.
- Corrected 10 Oct 2026: jev-agent-router, jev-mcp-router, jev-memory-selector and jev-effort-router leave "seen by name only"; the first three return Jev's answer to the agent (below). The Hermes catalog at
maindce1e9b(10 Oct, 03:23 UTC) has 13jeventries; it adds jev-cron-gate. - Leads triaged 10 Oct 2026: 35 leads from the 6–8 Oct scans in six classes (lead triage).
- Rechecked 10 Oct 2026: osuki 0.2.8 and the @jev-kit packages unchanged; @openclaw/typesafe 2026.9.9 published 8 Oct (OpenClaw had no counted row on 10 Oct). R12-S48, R12-S49
- Corrected 8 Oct 2026: on a Jev timeout, connection error, 429 or 402, Hermes core v2026.9.24 hands the approval question to the main agent model (if one is set), which can approve the command; other errors still escalate to a person (
auxiliary_client.pyL7610–7701, L7672–7675 atf97608f). jev-approvals (b) and (c):fail-closed→fallback-main agent model; "Tool call still runs?": No → Conditional. Documented R11-S110 to R11-S112; read at the pin 05:15–05:16 UTC, spot-checked against the raw file 05:21 UTC - Filled 8 Oct 2026: jev-curator (a) no key
fail-closed, (d) malformedfail-closed, (e) low confidenceheld(MIN_CONFIDENCE0.85); each was not recorded. Every cell of the 8 counted rows (on 8 Oct) is now recorded. Documented R11-S108, R11-S109; read once at the pin 05:15 UTC - Clarified 8 Oct 2026: rh-guard (e). The 0.5 is Jev's primary_kind confidence and was right;
REVIEW_THRESHOLD0.45 is a different measure, a floor on risk labels. Term (acts-anyway) and counts unchanged. Documented R11-S106 - Counts changed: on a Jev error, held or blocked 4/8 → 3/8, fallback to the main agent model 0/8 → 1/8, runs 4/8 (unchanged). No key or malformed: held 3/8 → 4/8, not recorded 1/8 → 0/8. Below threshold: held 3/8 → 4/8, not recorded 1/8 → 0/8.
jev-judge starts in shadow mode. After install it only logs what it would do. Its row describes enforce mode, which you switch on yourself. It is counted because the enforcing path is in the same repository.
ohmyjev's other pi and omp handlers also let the agent go on when Jev fails. They are not tool-call gates and are not separate rows. pi routing (before_agent_start, pi.ts L465–493): on a Jev failure no route is set and the current model is kept (L482); with no key, pi's own cheapest-model classifier routes up only (L486–490). pi done-check (agent_before_settle, L585–605): on a failure it does not block, and the agent settles (L594). omp routing (before_agent_start L682–703, classify L344–358): a failure sets no route (L354); with no key there is no routing (L348–351). omp done-check (session_stop L634–680): a failure does not block (L666–668). omp before_subagent_spawn (L714–725) makes no Jev call of its own; with no route the subagent's model is unchanged (L717). None of them blocks or switches the model on a Jev failure. Documented read at 78bb82c, 21 Oct 2026; not re-read at d307c39 or fbc4ea4 (R21-S101, R21-S102)
The 15 integrations that act: failure terms when Jev fails, and where the 29 come from
- Jev errors (timeout or HTTP error)3/15 fail-closed, 10/15 fallback, 2/15 fail-open; the tool call still runs in 5, conditionally in 2, not in 3, and 5 are routing rows (see the key finding)
- fail-closedjev-kit
- fail-closedbloudhood
- fallbackjev-approvals (main agent model)
- fail-closedjev-curator (guard mode)
- fallbackrh-guard (lexical detectors)
- fallbackopencode-tool-gate (opencode permission rules)
- fallbackosuki (default tier)
- fail-openjev-judge (enforce)
- fail-openohmyjev (pi, omp)
- fallbackjev-effort-router (configured model)
- fallbackjev-gateway (agent's own model)
- fallbackjev-triage (main model)
- fallbackjev-for-all launcher (default model)
- fallbackhermes-jev-mana (original Hermes branch)
- fallbackpi-jev-router (standard model)
- No API key5/15 fail-closed, 8/15 fallback, 2/15 fail-open
- fail-closedjev-kit
- fail-closedbloudhood
- fail-closedjev-approvals
- fail-closedjev-curator (guard mode)
- fallbackrh-guard (lexical detectors)
- fallbackopencode-tool-gate (opencode permission rules)
- fallbackosuki (default tier)
- fail-openjev-judge (enforce)
- fail-openohmyjev (pi, omp)
- fallbackjev-effort-router (configured model)
- fail-closedjev-gateway
- fallbackjev-triage (main model)
- fallbackjev-for-all launcher (default model)
- fallbackhermes-jev-mana (original Hermes branch)
- fallbackpi-jev-router (standard model)
- A malformed answer4/15 fail-closed, 9/15 fallback, 2/15 fail-open
- fail-closedjev-kit
- fail-closedbloudhood
- fail-closedjev-approvals
- fail-closedjev-curator (guard mode)
- fallbackrh-guard (lexical detectors)
- fallbackopencode-tool-gate (opencode permission rules)
- fallbackosuki (default tier)
- fail-openjev-judge (enforce)
- fail-openohmyjev (pi, omp)
- fallbackjev-effort-router (configured model)
- fallbackjev-gateway (agent's own model)
- fallbackjev-triage (main model)
- fallbackjev-for-all launcher (default model)
- fallbackhermes-jev-mana (original Hermes branch)
- fallbackpi-jev-router (standard model)
- Jev is unsure (below threshold)counted apart from errors: 4/15 held, 4/15 acts-anyway, 7/15 fallback
- Heldjev-kit
- Heldbloudhood
- Heldjev-approvals
- Acts anywayrh-guard (steer only)
- Acts anywayopencode-tool-gate
- Fallbackosuki (default tier)
- Acts anywayjev-judge (enforce)
- Acts anywayohmyjev (pi, omp)
- Heldjev-curator (below 0.85)
- Fallbackjev-gateway (agent's own model)
- Fallbackjev-effort-router (deepseek-v4.1-flash, medium)
- Fallbackjev-triage (main model)
- Fallbackjev-for-all launcher (default tier)
- Fallbackhermes-jev-mana (original Hermes branch)
- Fallbackpi-jev-router (standard model)
- Who publishes the 29by publisher
- Host vendor: 1/29OpenClaw
- TypeSafe: 0/29not found among the 29 read
- Community: 28/29GitHub repositories and npm packages
The 10 not counted: they do not act on Jev's answer themselves
These integrations were read in source but do not decide the tool call. Of the 10, 8 return Jev's answer to the agent and 2 give advisory hints. jev-approvals and jev-curator, not settled on 6 Oct, were settled on 7 Oct and moved to the counted table; jev-triage and hermes-jev-mana, not settled on 10 Oct, followed on 11 Oct. Hoppin-Sorter/jev-router, read on 11 Oct, and auto-mode and jcode, read on 21 Oct, are not settled and are not in these 10; nor is acmelabs-15/jev-gate, read on 30 Oct and not settled. Documented source at the pins in Sources, 6 Oct 2026; OpenClaw spot-checked against the raw files at the pin; the other 9 were read once at the pin
| Group | Integration (pin) | Hosts and hook point | Publisher | Why not counted | (a) No key | (b) Timeout | (c) HTTP error | Tool call still runs? |
|---|---|---|---|---|---|---|---|---|
| Returns to the agent | @openclaw/typesafe 2026.9.8npm (3 Oct 2026); provenance commit aa6008a | OpenClaw; decision provider; disabled by default | Host vendor (OpenClaw). npm provenance from the openclaw/openclaw repository; LICENSE "OpenClaw Foundation"; README "Official external plugin" Reported | Gives Jev to the agent's model through OpenClaw core's decision_evaluate tool (core not read) | no-decision"credentials-unavailable" | no-decision30,000 ms default (1,000–60,000) | no-decision | not applicableGates nothing; the agent sees "unavailable" |
| Returns to the agent | ourines/hermes-jev jev v0.1.2Catalog commit 28a4ea3 | Hermes Agent; jev_evaluate tool | Community (ourines); MIT | Tool only; catalog card: "advisory judgments, no automatic tool gating" | not recorded | not recorded30 s | Error returned to the agent | not applicable |
| Returns to the agent | ajensenwaud/hermes-jev-plugin jev-typesafe v0.1.0b3d29f7 (= catalog commit) | Hermes Agent; 4 plugin tools | Community (ajensenwaud); MIT | Tool only | no-decisionRaises | not recorded30 s | no-decisionError JSON to the agent | not applicable |
| Returns to the agent | ydmw74/jev-skill-router jev-skill-router-mcp76ee09e (= catalog commit) | Hermes Agent; MCP tool skill_select | Community (ydmw74); MIT per plugin.json, no LICENSE file | Tool only | no-decision | not recorded60 s | no-decision | not applicable |
| Returns to the agent | NiazMorshed2007/jev-review57690af | Codex, Cursor, OpenCode (also Claude Code); MCP tool jev_review | Community; MIT | Quality scores returned to the agent | no-decision | not recorded30,000 ms, up to 2 retries | no-decision | not applicable |
| Returns to the agent | burnigtm/jev-mcp v0.1.09448f61 | Codex, Cursor; MCP tools including gate | Community; MIT | gate returns auto, review or escalate; the CLI exits non-zero only on errors, so nothing is enforced | no-decision | not recorded30,000 ms | no-decision | not applicable |
| Returns to the agent | minhgv/jev-mcp v0.3.01a1f0a5 | Cursor, Codex; MCP tools plus a ci-shadow CLI | Community; MIT | ci-shadow prints would_block but always exits 0 | no-decision | not recordedNo explicit timeout; SDK default | no-decision | not applicable |
| Returns to the agent | pedroknigge/mcp_jev v0.0.1132377f8 | Cursor, Codex, Claude, Grok, Antigravity; MCP server | Community; MIT | "The MCP returns signals only — harness allowlist and sandbox still required" | no-decision | not recorded20,000 ms | no-decision | not applicable |
| Advisory hint | DECRUX9812/typesafe-skill-routerCatalog commit e6cdac2 | Hermes Agent; pre_llm_call hook | Community (DECRUX9812); MIT | advisoryAdds a skill hint; the agent's model still picks. Off by default | no-decisionSilent | no-decision10 s | no-decision | not applicable |
| Advisory hint | wellkilo/codex-jev-preflight633ddef | Codex; UserPromptSubmit hook adding context | Community; MIT | advisoryAdds a task-complexity note; gates no tool | not recorded | not recorded15 s | fail-openAdds an "unavailable" note and continues | not applicable |
How the 10 split: 8 return Jev's answer to the agent (@openclaw/typesafe, ourines jev, jev-typesafe, jev-skill-router-mcp, jev-review, burnigtm/jev-mcp, minhgv/jev-mcp, mcp_jev), so the agent's model decides what to do with it. 2 give advisory hints (typesafe-skill-router, codex-jev-preflight). 8 + 2 = 10 not counted; with the 15 counted rows and the 4 not settled, 29 read. DoGMaTiiC's jev-skill-router shares jev-judge's repository and is listed with it, not as a 19th integration.
Also returning to the agent, from the catalog and README only (not in the 24): Pinutss jev-agent-router (8cb6d67), jev-mcp-router (6a7b3ba) and jev-memory-selector (87bd9be). Each is one stdio MCP tool that "Registers no Hermes tools or hooks", so each returns Jev's answer to the agent; not a decision. Reported Hermes catalog at f97608f and the READMEs, 10 Oct 2026, 05:12–05:13 UTC; R12-S212 to R12-S218
Read in source, not settled: 3 integrations
Hoppin-Sorter/jev-router was read at a pin on 11 Oct 2026, 05:21 UTC, and auto-mode and jcode on 21 Oct 2026, 05:12–05:16 UTC (auto-mode re-read at 4.0.0 on 9 Nov 2026); none is counted either way until the open part can be read. A 4th not-settled integration, acmelabs-15/jev-gate, read on 30 Oct 2026, is in the next section. jev-triage and hermes-jev-mana, listed here on 10 Oct, were settled on 11 Oct and moved to the counted table (jev-triage, hermes-jev-mana). The jev-for-all OpenCode plugin is also not settled; it is described in the jev-for-all row, which is counted on its launcher. Cells use the shared vocabulary; they are not counted. Documented R13-S225 to R13-S228; 21 Oct: R21-S115 to R21-S138 (auto-mode), R21-S230, R21-S231 (jcode); 9 Nov: R29-S200 to R29-S207, R29-S17 (auto-mode at 4.0.0); read once at the pin
| Integration (pin) | Host and what it does | Why not settled | Error or timeout | Malformed | No key | Below threshold | Bypass |
|---|---|---|---|---|---|---|---|
Hoppin-Sorter/jev-router 0.7.0bdb4721 (plugin.json 0.7.0). Read 11 Oct 2026 (was README only) | Claude Code (Codex and Cursor appear only as agents that use the router outside Claude Code; not read). In its default "auto" mode it rewrites model (and effort if effortRouting is on; default off) on every main-loop request in a turn.step middleware (hooks/register.tsx L484–497) | Not settled: host handling of the model rewrite not readable at a pin (the Claude Code engine source is not public) | fallback-previous pick or session model3,000 ms; keeps the previous pick if active within 10 min, otherwise the session model (L427–431, L490–496) | fallback-previous pick or session modelAs on error (L170–175); a missing tier answer gives the default tier claude-sonnet-5-5 (L410, L437–447) | fallback-session modelStatus "Jev: no API key" (L420–425) | acts-anywayPicks the cheapest tier that clears a cumulative threshold, else a higher one, up to "deep" (claude-opus-5-5) (lib/jev-router.ts L134–139, L152–159) | Recorded: / commands; non-composer origins; turnId set; mode off or shadow; subagents keep their own model (L395–404, L487–490) |
zgeoff/auto-mode 4.0.0 (Claude Code mod)7b246a7 (tag auto-mode-v4.0.0; npm 8 Nov 2026). Read 9 Nov 2026 at 4.0.0: failure path unchanged (cited files identical to 3.1.6, 382b7a9); bypass now lists 7 skip stages, adding a reply-channel allow (present since 3.1.x); still not settled because the mod host's handling of the tool.check answer was not read. 4.0.0's own changes are in src/judge/*, a review judge that runs only after a deny; its CHANGELOG: "a Messages API judge now runs the claude on the PATH, so a machine without claude fails its reviews" Reported (R29-S200 to R29-S207). Earlier pin: c0b9378 (tag 3.0.0, 20 Oct 2026); read at 2.0.2 (458c41c) and the cited files re-read at 3.0.0: gate path unchanged. Read 21 Oct 2026 (was a lead). Checked 30 Oct 2026, 05:20 UTC: auto-mode 3.1.5 (29 Oct 2026); a comparison with 3.0.0 found no change to its Jev call or failure handling in the files compared (a targeted comparison of the Jev path, not a full re-read); it still requests jev-1.13.0 by ID (R24-S37) | Claude Code, as a mod. Its code answers Claude Code's permission question with allow or deny in place of "ask"; it acts only when Claude Code would ask (register.ts L118–120, L198). Default model jev-1.13.0 (presets.ts L15) | Not settled: how the mod host handles the tool.check answer was not read in the host's own source, and the stages that run before Jev were named but not read line by line | held for the user's prompt (as read, not counted) 5,000 ms; on an error no verdict is written and Claude Code's question stays (default onFailure: 'defer', config.ts L76) | held for the user's prompt (as read, not counted) only exact allow or deny answers are used (parse-decision.ts L3–33) | held for the user's prompt (as read, not counted) the call throws and the question stays (classify-with-jev.ts L51–54) | acts-anyway (as read, not counted) blocks only at a block probability ≥ 0.2 (config.ts L69, L77); below it the call runs without asking | Partial read: 7 skip stages are listed (classify-action.ts L234–242; 6 at 3.0.0), including a reply-channel allow present since 3.1.x (classifyReplyChannel, L240; L470–488: a tool call matching a reply channel in the task scope is allowed with deciding stage "bypass"); their internals were not read; not counted as recorded |
1jehuang/jcode, PR #1890 "Make Jev decision models configurable"merge f5c4922 (19 Oct 2026). Read 21 Oct 2026 (was a lead) | jcode (its own coding agent). Adds agents.memory_jev_model, agents.browser_jev_model and agents.voice_jev_model (and JCODE_*_JEV_MODEL). Jev picks which memories are injected (memory.rs L1181; on an error nothing is injected, L1190), a browser action (browser_jev.rs L2) and a voice intent (voice_intent.rs L193–199) | Not settled: no tool-call gate or model routing was found in the files read; its other callers (TUI, provider-core, overnight-core) were not read | not recorded (memory injection: nothing injected on an error) | not recorded | not recorded | not recorded | not recorded |
Read in source; not on the failure map yet
acmelabs-15/jev-gate was read at a pinned commit of 20 Oct 2026 on , 05:21–05:28 UTC (five failure cells, spot-checked against the raw files at the pin). It is not settled, so it is not counted either way, and it is not on the failure map (the map still counts 80). With the 3 rows above it makes 4 not settled of the 29 read in source. Its cells use the shared vocabulary and the colours of the legend on this page; they are not added to any count of the counted table, the figure or the map. Documented (source at pin; its README's claims are Reported) R24-S228 to R24-S234, R24-S43; read once at the pin
Key finding (Documented, 30 Oct 2026; source read at a 20 Oct 2026 pin): acmelabs-15/jev-gate, a Claude Code hook, lets the tool call go ahead when Jev errors, times out after 1.5 seconds, answers malformed or has no key; by default it only logs. One integration, five cells at one pin. Claude Code's own handling of the hook's answer was not read in source, and nothing here says whether anyone uses it.
@caph42/pi-jev-router 2.0.0, read at 75ad5b7 with pi 1.5.0 (f9a2636), . It registers a virtual model jev/auto; pi sends each request to the model the router picks. When Jev times out (3 s), errs, answers badly or has no key, the request goes to the standard model. Some turns skip Jev (retries, continuations, compaction summaries). It is counted (a routing row: the turn runs on another model, and the turn's tool calls run on it) and is in this page's totals and figure (29 read, 15 counted, 10 fallback on a Jev error, 5 routing rows); it is not on the failure map, which still counts 80. Its cells use the same terms and legend colours as the counted table. Documented source at the pins; jev-router.ts L118, L453, L467, L485–493 spot-checked against the raw file at the pin, 8 Nov 2026, 05:27 UTC; R28-S200 to R28-S203, R28-S205 to R28-S208, R28-S210, R28-S48; read once at the pin; the fallback models were not read
| Integration (pin) | Host and what it does | Counting verdict and why | Error or timeout | Malformed | No key | Below threshold | Bypass |
|---|---|---|---|---|---|---|---|
acmelabs-15/jev-gate (@acmelabs/jev-gate 0.2.1)b3ae111 (main, 20 Oct 2026, 13:38 UTC). Read 30 Oct 2026 | Claude Code plugin: a PreToolUse hook on Bash, Write, Edit, MultiEdit, NotebookEdit and WebFetch (hooks/hooks.json L4–14); host hook timeout 5 s. Jev call: scripts/gate/hook.ts L116 → scripts/gate/jev.ts L30 POST /v1/systemone; default model jev-latest (config.ts L65). Default mode shadow: logs only (config.ts L66); enforce mode outputs permissionDecision ask (default) or deny (hook.ts L119, L131) | Not settled: the host path was not read in source (Claude Code's handling of permissionDecision is not public source; only its hooks reference was read, R24-S234). With the default ask a person decides; the default shadow mode is log-only | fail-openTimeout 1,500 ms (jev.ts L37, config.ts L69); the catch logs and returns nothing (hook.ts L132–141, L149) | fail-openNo answers object throws into the same catch (jev.ts L44–47); a missing field counts as 0 (jev.ts L56–62, L69) | fail-openSkipped and logged (config.ts L76–78, hook.ts L91–100) | acts-anywayNormal permission flow; it flags only at >= 0.9 (jev.ts L69, config.ts L68) | recordedTools outside the matcher (for example Read, Grep and MCP tools) are never scored (hooks.json L6); unparseable hook input (hook.ts L74–77); no key. Entry points read: 5 (hook.ts, session-start.ts, two SKILL.md files, report.ts) |
Caph-dev/pi-jev-router (@caph42/pi-jev-router 2.0.0)75ad5b7 (npm gitHead of 2.0.0, published 1 Nov 2026; extensions/jev-router.ts in the npm tarball is byte-identical to the file at the pin); host pi 1.5.0 at f9a2636 (tag v1.5.0). Not r33drichards/pi-jev-router, a different project. Read and counted 8 Nov 2026 | pi extension that registers a virtual model jev/auto. Host path read at the pin: registerVirtualModel stores the route (model-runtime.ts L972–980); resolveModel calls it and returns the physical model (L1011–1041); each agent-loop request is sent with model: route.model (agent-session.ts L830–858), so pi acts on the choice by code. Jev call: classify on typesafe/jev-latest (jev-router.ts L459–467; L134, L434), sent to api.typesafe.ai/v1/systemone through pi-ai. Mode: routes a model choice per request (P(complex) ≥ 0.5 → the complex model, else the standard model; guarded handoff to an implementation model at ≤ 0.2 after the first successful edit or write); an opt-in record strategy only logs (L1160–1169) | Counted, a routing row: it reroutes a model choice and pi enforces it by code (host path read at a pinned commit). Community (Caph-dev). Counted on this page; the failure map's 80 is unchanged | fallback-standard model3,000 ms default (L118, L439, L453), no retry (L467); the standard model on the first routed turn, otherwise the current model is kept; the request goes ahead (L485–486, L490) | fallback-standard modelAn invalid answer takes the same path (L491–493); pi's adapter turns a bad body into an error result (pi system-one-shared.ts L230–235) | fallback-standard modelPer request; no startup error. pi throws "Provider is not configured" (model-runtime.ts L677), caught into an error result (L829–830), then L490 | standard modelP(complex) below 0.5 goes to the standard model; no separate confidence cut (L135, L197, L1055); a downgrade from the complex model only at ≤ 0.15 after 2 turns (L199–201) | recordedTurns routed without Jev: direct requests such as compaction summaries (L992), retries (L1032–1046), a reused planning model (L1049–1053), continuation turns and already-evaluated user turns (L1067–1113), classifier off or empty input (L431–432), every turn after the first under fixed (L1182). Entry points from one listing: default export (/jev-router command L870, session_start L915–918, route() L929), DEFAULT_CONFIG L115, summarizeClassifierUsage L288, loadConfig L820, lib/diagnostics.ts helpers |
Lead triage: 35 leads from the 6–8 Oct scans, 31 lines from the 8–11 Oct scans, 13 lines from the 19–20 Oct scans, 10 lines from the 21–30 Oct scans and 19 lines seen 24 Oct–8 Nov 2026
Unique leads from daily scans on 6–8 Oct, sorted into six classes from their README or package description on 10 Oct 2026, 05:13 UTC. Not a survey; nothing here is counted unless a row above says so. Reported R12-S220 to R12-S253; dsh-jev-verify and pi-system-one appeared in both scans and are counted once
- 35 leads by class11 + 9 + 6 + 5 + 3 + 1 = 35
- Candidate gate: 11could decide a tool call or model choice
- Claude Code: 9for Claude Code and MCP guards
- Returns to the agent: 6not a decision
- Not a coding-agent integration: 5
- In-host shaping: 3skill or context selection inside the host
- Not readable: 1HTTP 404
- The 11 candidate gates5 read at pins, 6 README only (11 Oct 2026): 4 + 1 + 6 = 11
- Read, counted: 4jev-gateway (10 Oct); jev-triage, hermes-jev-mana, jev-for-all launcher (11 Oct)
- Read, not settled: 1Hoppin-Sorter/jev-router
- README only: 6not counted
- 31 lines from the 8–11 Oct scans by class10 + 7 + 8 + 3 + 2 + 1 = 31 lines; 30 unique projects
- Candidate gate: 10could decide a tool call or model choice
- Claude Code: 7for Claude Code and MCP guards
- Returns to the agent: 8not a decision
- Not a coding-agent integration: 3
- In-host shaping: 2
- Not readable: 1no HEAD
The 6 candidate gates from the 6–8 Oct scans still read from README only: pi-safely, pi-jev-classifier, dsh-plugin-jev-approval, dsh-jev-verify, Astra-Ares-Multi-Agent and @cr1ms0n/pi-subagent. Their source was not read, so their failure behaviour is not recorded. Changed 11 Oct 2026: 8 → 6; jev-for-all (counted on its launcher) and Hoppin-Sorter/jev-router (not settled) were read at pins
The 8–11 Oct scan lines (lines 36–66 in the table below) were sorted on 11 Oct 2026, 05:21–05:23 UTC, from README or package description only, into the same six classes. The 20 items from the 10 Oct 2026 scan: candidate gate 7, Claude Code 4, returns to the agent 5, not a coding-agent integration 2, in-host shaping 2, not readable 0 (7 + 4 + 5 + 2 + 2 + 0 = 20). The 11 items from the 11 Oct 2026 scan: candidate gate 3, Claude Code 3, returns to the agent 3, not a coding-agent integration 1, in-host shaping 0, not readable 1 (3 + 3 + 3 + 1 + 0 + 1 = 11). Lines 40 and 66 are the same project (eachann1024/pi-jev-route and its npm package), so 31 lines are 30 unique projects. None of the 31 was counted on 11 Oct; line 51 (ohmyjev) is counted since 21 Oct as the pi and omp row. Reported R13-S229 to R13-S259; pins are HEADs by git ls-remote
All 66 lines with host, pin and class (35 leads from 6–8 Oct, 31 lines from 8–11 Oct)
| # | Repository or package | Host | Pin or version | Claimed behaviour (Reported) | Class |
|---|---|---|---|---|---|
| 1 | PeterFujiyu/jev-status-marketplace | Claude Code | dd98512 | after-turn status band | Claude Code |
| 2 | xiaoxiaomugong/jev-context | other (CLI) | 0be6e1c | budgeted code-reading list | returns to the agent |
| 3 | apatlasdev/fast-decide-compaction | Claude Code | 88e4d85 | compaction via Mercury Decide, not Jev | Claude Code |
| 4 | dcowsill/pi-safely | pi | b42c752 | gates non-allowlisted tool calls (example failOpen: true) | candidate gate (README only; not counted) |
| 5 | RanyAlbegWein/jev-triage (counted 11 Oct) | OpenClaw | 6360e21 | triages inbound messages | counted 11 Oct candidate gate, read (B4) 10 Oct; host path read 11 Oct |
| 6 | matthewpetela/opencode-decisions-plugin | OpenCode | d8ace9f | decisions_evaluate tool | returns to the agent |
| 7 | abehrman/hermes-jev-mana (counted 11 Oct) | Hermes Agent | 3daf639 | core patch plus decision points | counted 11 Oct candidate gate, read (B4) 10 Oct; full patch read 11 Oct (inert after the documented install) |
| 8 | moisaporto/pi-jev-classifier | pi | 74fbd2e | blocks destructive, irreversible bash/write/edit | candidate gate (README only; not counted) |
| 9 | eriverjpn/dsh-plugin-jev-approval | other (DeepSeek Harness) | 962db8a | allow ≥ 0.8 / reject ≤ 0.2 / a person in between | candidate gate (README only; not counted) |
| 10 | SylTi/jev-context-loader | Codex / OpenCode / Hermes | aa79f34 | loads skills and MCP tools on demand | in-host shaping |
| 11 | romantcig/brave-jev-mcp | other (MCP) | 7ba1036 | filters search results | returns to the agent |
| 12 | jev-gateway (row) | Codex, OpenCode, Claude Code and 3 others | 0.5.1 | Jev picks the tool | candidate gate → read (B4): counted |
| 13 | @delt/claude-jev-advisor | Claude Code | 0.4.1 | compact advice plus an rm guard | Claude Code |
| 14 | pi-skill-picker-jev | pi | 0.3.0 | skill selection | in-host shaping |
| 15 | pi-system-one | pi | 1.3.4 | system_one tool plus /judge | returns to the agent |
| 16 | dsh-jev-verify | other (DSH) | 0.8.6 | tools plus opt-in autoGuard pre-execute deny | candidate gate (README only; not counted) |
| 17 | mcowdery/claude-model-guard | Claude Code | 8b6ade2 | model-tier block or nudge | Claude Code |
| 18 | VictorGambarini/jev-mod | Claude Code | c39353a | routing plus gates, off by default | Claude Code |
| 19 | Hoppin-Sorter/jev-router | Claude Code, Codex, Cursor | bdb4721 | per-prompt model and effort pick | not settled read 11 Oct at bdb4721 (see Read in source, not settled) |
| 20 | kalowery/jev-router | — | HTTP 404 | not recorded | not readable |
| 21 | DiegoSalazar/ask-jev-skill | Claude Code | e2e313d | skill plus CLI | Claude Code |
| 22 | kanedayo/clef-mcp | Claude Code | 91679b6 | MCP server for Cloudflare Clef, not Jev | Claude Code |
| 23 | Sad1mus/agentic-portfolio | Claude Code | a9cfa37 | demo MCP plus skills | Claude Code |
| 24 | NighCross/jev-netlify-mcp | Claude Code / Codex / any MCP client | 545a892 | proxy plus MCP server | returns to the agent |
| 25 | JixiangChen-Jimmy/Astra-Ares-Multi-Agent | Codex (patched) | 60dc127 | Jev picks effort and child model | candidate gate (model choice; README only) |
| 26 | huayaney-exe/gutfeel | other | cc91932 | usability testing for MCP servers | not a coding-agent integration |
| 27 | jev-for-all | OpenCode (+ Claude Code hook) | 0.7.1 | routes skill and tool choices | counted 11 Oct read at npm 0.7.1: start launcher counted (row); OpenCode plugin part not settled |
| 28 | @yadsh/dsh-jev-compaction | other (DSH) | 0.2.1 | prunes stale tool output | in-host shaping |
| 29 | jev-agent-tools | pi / omp / MCP | 0.4.0 | six evidence tools | returns to the agent |
| 30 | @cr1ms0n/pi-subagent | pi | 0.11.15 | Jev picks the subagent's model | candidate gate (model choice; README only) |
| 31 | @muthuishere/jevx | Claude Code, Codex | 0.12.13 | skill plus CLI; optional hooks | Claude Code |
| 32 | @scruple/provider-jev | other | 0.3.5 | provider for Scruple | not a coding-agent integration |
| 33 | @taifoon/jev | other | 0.2.18 | on-chain job grader | not a coding-agent integration |
| 34 | @lakeday/perch | other | 0.4.2 | semantic linting | not a coding-agent integration |
| 35 | jev-fabric | other | 0.6.0 | process orchestration | not a coding-agent integration |
| 36 | Krzysztof-Cieslak/jev-agent-control | OpenCode V2 | 43b26a5 | routes requests between Plan, Build and user agents; auto handoff | candidate gate (README only) |
| 37 | mrhsiung2015/jev-model-router | Codex (skill) | 1c75436 | the script outputs one JSON recommendation; Codex delegates per SKILL.md | returns to the agent |
| 38 | joslynSmall/pi-gear-router | pi | 76db798 | gears/auto virtual model; 3-tier model plus thinking level | candidate gate (README only) |
| 39 | qianyuxiang-369/pi-jev-governor | pi | 4d8cb24 | tier, plan and permission (human or fail-closed fallback), outcome | candidate gate (README only) |
| 40 | eachann1024/pi-jev-route | pi | f2f75f0 | subagent model selection; "Routing is enabled by default" | candidate gate (README only) |
| 41 | alexswan10k/pi-extensions | pi | 98434b6 | ask_jev tool (local LM Studio model) | returns to the agent |
| 42 | SaulineytorX/urdia | other (local agent workspace) | 9887249 | Jev selects the agent profile; explicitly enabled, manual by default | candidate gate (README only) |
| 43 | HunterXing/jev-judge | DSH / Claude Code / Codex / OpenCode | 97df122 | "changes only what the model sees" | in-host shaping |
| 44 | hakilee/paseo-jev-compaction | Codex in Paseo | 3450cd8 | Jev history compaction | in-host shaping |
| 45 | oozoofrog/jev-native | other (macOS computer use) | e88caaa | standalone native controller | not a coding-agent integration |
| 46 | TiChuXiXi/dsh-jev-zxh | DSH | 0f5d0ea | jev_decide tool | returns to the agent |
| 47 | lucaminudel/jev-unpack-hook | Codex and Claude Code (Stop hook) | 999a732 | blocks Stop to re-pace multi-ask messages | candidate gate (Stop-hook block, not a tool or model choice; README only) |
| 48 | 8itlab/reflex-cu | MCP (Claude Code, Codex) | 3887865 | computer-use MCP server with quick judgements | returns to the agent |
| 49 | Nancy97940/building-with-typesafe-jev | none | 94b6576 | link list | not a coding-agent integration |
| 50 | randlee/plan-anchor | Claude Code and Codex (skill) | 7c1f1ce | planning and development anchor gates run by the skill | returns to the agent |
| 51 | Novacon/ohmyjev | Claude Code, Codex, pi, omp | 6bb7934 | blocks irreversible commands; routes effort | candidate gate (README only) counted since 21 Oct 2026: pi and omp row |
| 52 | Oqura-ai/optim-jev | Claude Code | 212b702 | compaction (on), routing and skills (off) | Claude Code |
| 53 | Akurganow/model-router | Claude Code | 862e3f7 | subagent model and effort | Claude Code |
| 54 | dominicrico/jev-router | Claude Code | 8434b75 | model routing | Claude Code |
| 55 | cloud-f1/jev-agent-kit | Claude Code | 6a7aa8e | shortens Bash output; opt-in | Claude Code |
| 56 | murilogteixeira/opencode-jev-pre-llm | OpenCode V2 | bbe16ae | classifies prompts before the LLM; "injecting determinism detection" | returns to the agent |
| 57 | Misato-R/Codex_Model_Router | Codex (CLI router) | 6997343 | local rules (dry-run) by default; Jev optional; maps to Codex tiers | candidate gate (model choice; README only) |
| 58 | sunu-dev/jev-pilot | Codex CLI, Claude Code (launchers) | acddefa | per-turn model and effort routing | candidate gate (model choice; README only) |
| 59 | chenrobotcat/Model_Selection | Claude Code (PreToolUse) | 646b6a2 | subagent model via updatedInput; local Ollama by default | Claude Code |
| 60 | gecm0/jev-mod | Claude Code | 4f3f55e | $.jev library for other mods | Claude Code |
| 61 | gecm0/skill-router-mod | Claude Code | 33584c3 | skill suggestions; "the model decides" | Claude Code |
| 62 | iptton-ai/dsh-jev | DSH | 875a3ad | jev_evaluate tool | returns to the agent |
| 63 | adityajha2005/decisionkit | CLI / Claude Code plugin / MCP | 06da6f7 | provider-neutral decision API | returns to the agent |
| 64 | kevkzumi/jev_integration | none (prompt notes) | 2b9a356 | prompts to paste into agents | not a coding-agent integration |
| 65 | Smyan1909/hunchcraft | — | no HEAD (05:22 UTC) | not recorded | not readable |
| 66 | npm @each1024/pi-jev-route | pi | 1.0.6 | subagent model selection (the npm package of line 40) | candidate gate (model choice; README only) |
The 20 leads from the 8–10 Oct scan (R12-S02), listed here untriaged on 10 Oct, are lines 36–55 of the table above since 11 Oct 2026.
Lead triage, 19–20 Oct 2026: 13 lines
13 leads classed from one file each at a pin on 21 Oct 2026, 05:12–05:20 UTC (lines 67–79). 5 act on Jev's answer in a host plugin or agent and are candidates for a later full read; 1 acts in its own test setup; 1 acts only on context compaction; 2 return the answer to the agent; 1 is a client library; 2 call another model, not Jev; 1 is a feature request with no code (5 + 1 + 1 + 2 + 1 + 2 + 1 = 13). None of the 13 is counted: a counted row needs all five failure cells read at a pin. Documented for the lines cited; Reported for anything beyond them R21-S200 to R21-S227
- 13 lines by class5 + 1 + 1 + 2 + 1 + 2 + 1 = 13
- Acts on Jev's answer: 5candidates for a later full read
- Own test setup: 1recoveriq
- Compaction only: 1pi-typesafe-compact
- Returns to the agent: 2not a decision
- Client library: 1no decision
- Not Jev: 2another model
- Feature request: 1no code
The 13 lines with host, pin, file and class (19–20 Oct scans)
| # | Repository or package | Host | Pin or version | File and line read | Class |
|---|---|---|---|---|---|
| 67 | chinmayakelkar/jev-plan-review | Claude Code (PreToolUse, ExitPlanMode) | 894f673 | src/hook.ts L103–110 denies ExitPlanMode; Jev call src/jev.ts L146 | acts on Jev's answer: candidate for a full read |
| 68 | menoncello/model-selector | Claude Code (PreToolUse, Agent and Workflow) | 63eddf9 | model-router/router.ts L41, L52–53; Jev call model-router/jev.ts L32 | acts on Jev's answer: candidate for a full read (model choice) |
| 69 | Covre81/jev-router | Claude Code (local gateway) | d83e068 | src/routing/router.ts L259, L274, L279; Jev call src/classifier/jev-classifier.ts L87 | acts on Jev's answer: candidate for a full read (model choice) |
| 70 | Keen Code (mochow13/keen-code) | Keen Code (its own agent) | c77a3a6 | internal/cli/repl/router.go L283, L290, L329; Jev call internal/decision/typesafe/evaluator.go L120 | acts on Jev's answer: candidate for a full read (model choice; opt-in) |
| 71 | npm pi-jev-auto-mode 0.6.0 | pi | gitHead b4bf6c0 | src/extension.ts L291, L957–960; Jev call src/jev/engine.ts L95 | acts on Jev's answer: candidate for a full read (tool-call gate) |
| 72 | sanghamitrajohri/recoveriq | its own test setup (a coding-agent loop with pytest) | a8241df | strategies.py L85–105; Jev call jev_adapter.py L138 | acts on Jev's answer in its own test setup (recovery action); not a host plugin |
| 73 | npm @narumitw/pi-typesafe-compact 0.1.2 | pi | npm 0.1.2 (no gitHead) | src/typesafe-compact.ts L136–141, L278; Jev call src/evaluator.ts L130 | acts on Jev's answer for context compaction only; decides no tool call or model choice |
| 74 | npm @narumitw/pi-typesafe 0.1.2 | pi | npm 0.1.2 (no gitHead) | src/jev.ts L120; src/client.ts L154, L219–220 | returns to the agent |
| 75 | npm @narumitw/pi-typesafe-search 0.1.2 | pi | npm 0.1.2 (no gitHead) | src/typesafe-search.ts L44; src/jev-client.ts L133 | returns to the agent (reranks the search results it returns) |
| 76 | npm pi-typesafe-ai 0.2.1 | pi | gitHead a87b954 | src/client.ts L55; src/index.ts L34 | client library: returns to the caller; no decision |
| 77 | Mrmimee/opencode-v2-jev | OpenCode V2 | ea576ca | jev.ts L6–7, L238 | not Jev: calls an Agnes model |
| 78 | Mrmimee/codex-jev | Codex | da8e784 | plugins/jev/jev_engine/__init__.py L64–65 | not Jev: Agnes backend (a port of the same account's hermes-plugin-jev) |
| 79 | openchamber issue #5014 | OpenChamber | issue (open; no linked pull request) | issue page | feature request, no code |
Lead triage, 21–30 Oct 2026: 10 lines
10 leads from the 21–30 Oct 2026 scans (lines 80–89), each classed on , 05:23 UTC, from its README and at most one listing at a pin or package version. Lead, not read: their source was not read, none is counted and their failure behaviour is not recorded. 5 are candidate gates (one of them for model choice), 1 likely returns to the agent, 1 is an advisory suggestion card, 1 is not a coding-agent hook, 1 is archived and 1 is a write-up with no repository (5 + 1 + 1 + 1 + 1 + 1 = 10). Reported R24-S235 to R24-S245
- 10 lines by class5 + 1 + 1 + 1 + 1 + 1 = 10
- Candidate gate: 5README only; candidates for a later full read
- Likely returns to the agent: 1Unverified
- Advisory: 1suggestion card
- Not a coding-agent hook: 1document triage
- Archived: 1moved
- Write-up: 1no repository
The 10 lines with host, pin or version and class (21–30 Oct 2026)
| # | Repository or package | Host | Pin or version | Claimed behaviour (Reported) | Class |
|---|---|---|---|---|---|
| 80 | sanketmuchhala/jevgate | not a coding agent | b344c8e (26 Oct 2026) | Document triage service | lead, not read: not a coding-agent hook R24-S235 |
| 81 | npm pi-typesafe | pi | 0.13.0 (24 Oct 2026; latest 0.13.1) | pi evaluation tool and typed API | lead, not read: likely returns to the agent, not a decision (Unverified) R24-S236 |
| 82 | pbswimmer3/claude-jev-router | Claude Code | be11bad (24 Oct 2026) | Advisory model and effort suggestion card | lead, not read: advisory R24-S237 |
| 83 | hasanzaman22/jev-effort-router | Claude Code (plugin) | 28eea82 (21 Oct 2026) | Sets the effort per prompt | lead, not read: candidate gate (model choice) R24-S238 |
| 84 | G0-0000/subagent-jev | CodeBuddy Code, Claude Code and pi | 6c79c42 (24 Oct 2026) | Subagent-dispatch gate; README says it "fails open on errors" | lead, not read: candidate gate R24-S239 |
| 85 | Pyromind-Dynamics/pi-jev-governor | pi | 5b00d98 (21 Oct 2026) | Governance layer (planning, routing, permission control); README says "fail-closed" | lead, not read: candidate gate R24-S240 |
| 86 | Komugi-powerplatform/yomiyasu-sem | Claude Code (Stop hook) | 5a81d35 (22 Oct 2026) | Stop-hook layer that sends spec edits back | lead, not read: candidate gate R24-S241 |
| 87 | ysys143/jev-harness | not stated | e716896 (22 Oct 2026) | Archived; README "Moved to context-harness" (successor not opened) | lead, not read: archived R24-S242 |
| 88 | felinae98/cmd-mod-jev-guard | Command Code | 4051a8b (21 Oct 2026) | Tool-call guard answering deny, ask or allow | lead, not read: candidate gate R24-S243 |
| 89 | Parallix review-loop article (HN 50205651) | not stated | article dated 21 Oct 2026 | Personal workflow write-up | lead, not read: write-up; no repository, not a graded comparison R24-S244, R24-S245 |
Coding-agent projects seen 24 Oct–8 Nov 2026 (lead, not read in full)
19 projects seen in the 24 Oct–8 Nov 2026 scans, each classed on , 05:18–05:19 UTC, from its README only (no listing opened). Lead, not read in full: their source was not read, none is counted, no count on this page changes and their failure behaviour is not recorded. By host: Codex 6, OpenCode 4, pi 4, Claude Code 5. By class: 8 candidate model-choice, 1 candidate gate, 2 return to the agent, 3 advisory, 5 not a coding-agent integration (8 + 1 + 2 + 3 + 5 = 19). Quotes are from each README; translations in brackets are ours. 73 further names seen in the same period were not triaged (no list). Reported R28-S122 to R28-S129, R28-S131 to R28-S141
- stevenke1981/jev: Codex; candidate model-choice (a local proxy that rewrites the reasoning-effort field); README read , first seen 24–30 Oct 2026; "Jev 是夾在 Codex 與上游之間的本地 proxy,只改寫
reasoning.effort一個欄位" [our translation: Jev is a local proxy that sits between Codex and the upstream and rewrites only one field, reasoning.effort]; lead, not read in full R28-S122 - RAYDO-MAX/codex-model-advisor: Codex; advisory (a skill recommends model and effort; the user switches); README read , first seen 24–30 Oct 2026; "It does not execute the original task in that response or automatically switch the active model."; lead, not read in full R28-S123
- mxttbennett/jev-relayer: Codex and Claude Code; candidate model-choice (per-prompt placement, model and effort; "It never blocks a prompt"); README read , listed 4 Nov 2026; "picks the model, and the reasoning effort, for that turn"; lead, not read in full R28-S124
- elizabethsiegle/jev-entire-demo: Codex (demo); not a coding-agent integration (an offline demo script that asks Jev about a recorded session; no hook); README read , listed 6 Nov 2026; "Jev checks the request, session and diff for evidence that the tests ran."; lead, not read in full R28-S125
- AJackTi/antigravity-jev-skills: Codex / Antigravity; returns to the agent (a skill; Jev triages evidence for the agent); README read , first seen by 2 Nov 2026; "
jev: typed decisions and evidence triage."; lead, not read in full R28-S126 - VishalJ99/lock-in: Codex (observed app); not a coding-agent integration (a macOS playground comparing Decisions and Jev on Codex activity); README read , first seen by 2 Nov 2026; "A small macOS app for comparing Decisions and Jev against activity in Codex and Chrome."; lead, not read in full R28-S127
- jflowers/opencode-auto-question: OpenCode; returns to the agent (the plugin answers the agent's
questiontool in place of the human; its README says failures fall back to the human form); README read , first seen 24–30 Oct 2026; "an AI decision engine answers on behalf of the human"; lead, not read in full R28-S128 - crisesarmiento/opencode-routing-dashboard: OpenCode; not a coding-agent integration (a local dashboard over recorded router logs); README read , listed 4 Nov 2026; "Jev suggestions, recorded context agents, route decisions, aggregate model usage and errors."; lead, not read in full R28-S129
- ezcodelab/libreharness: OpenCode; advisory (README: Jev "never the deciding gate"); README read , listed 4 Nov 2026; "Jev, a first-class advisory layer that asks typed questions about a diff"; lead, not read in full R28-S131
- AleksandrBagdasarov/opencode-jevcode: OpenCode (v2 plugin); candidate model-choice (a
jevCodeagent whose coding model per turn Jev picks; its README claims fail-open); README read , listed 3 Nov 2026; "If Jev is slow, down or not configured,jevCodebehaves like a plain agent"; lead, not read in full R28-S132 - miketropi/pi-jev-model-manager: pi; candidate model-choice (virtual model
jev-manager/auto); README read , first seen 24–30 Oct 2026; "A reference Pi extension for automatic model choice via TypeSafe Jev with a native management UI."; lead, not read in full R28-S133 - hsmoviich/omp-maestro: pi (omp); candidate model-choice (a bundle with its own "Jev Router"); README read , first seen 24–30 Oct 2026; "sent to the smallest model that can do it well. Fails open, never down."; lead, not read in full R28-S134
- Paul-Yuchao-Dong/pi-omarchy-cu: pi; not a coding-agent integration (a computer-use extension; Jev picks a desktop UI target, not a tool call or model); README read , first seen by 2 Nov 2026; "asks Jev System One for the target Ref and Operation"; lead, not read in full R28-S135
- ttdua/pi-jev-pilot: pi; candidate model-choice (subagent model tier and effort); README read , listed 5 Nov 2026; "Each eligible
subagentlaunch gets a model tier and thinking level chosen from how hard the task is, overriding what the main model asked for."; lead, not read in full R28-S136 - alejandroparis17/router-jev: Claude Code (dispatches to Codex / Antigravity); candidate model-choice (
UserPromptSubmitrouting; also aPreToolUseguard, a possible gate lead); README read , repository created 7 Nov 2026, 18:58 UTC; "guard.js (PreToolUse hook; a Stop hook only records) makes the main session delegate instead of doing the work itself"; lead, not read in full R28-S137 - JingYao152/jev-claude-gate: Claude Code; candidate gate (four hooks including a
PreToolUsegate); README read , first seen 24–30 Oct 2026; "PreToolUse Hook: 工具与命令执行门禁 (jev-gate.py)" [our translation: PreToolUse hook: tool and command execution gate (jev-gate.py)]; lead, not read in full R28-S138 - sunjongos/jev-typesafe-gatekeeper: Claude Code (skill); not a coding-agent integration (a skill that triages inbound data such as customer-service and clinical inquiries); README read , first seen 24–30 Oct 2026; "초경량·초고속 판단(Decision & Classification) 전담 AI 에이전트 스킬" [our translation: an ultra-light, ultra-fast AI agent skill dedicated to judgement (Decision & Classification)]; lead, not read in full R28-S139
- tbmsindemnify/claude-optimize-mod: Claude Code (mod); candidate model-choice (Jev routing optional, off without a key); README read , first seen by 2 Nov 2026; "Picks the model, effort and skill with Jev."; lead, not read in full R28-S140
- NOOB-3301/claude-code-evolve: Claude Code; advisory (Jev gates whether a Haiku note-writer runs; notes are added to later prompts; no tool-call or model-choice gate); README read , first seen by 2 Nov 2026; "Turn gate: Jev answers a yes/no
worthquestion per turn. Below the threshold (0.5), Haiku never runs."; lead, not read in full R28-S141
Kilo Code: no Kilo Code Jev integration found on GitHub repository search ("kilocode jev") or npm search ("kilo jev"), 21 Oct 2026, 05:12 UTC. R21-S228, R21-S229
Install-and-check list
Six questions to answer before you rely on any of these. Documented from the rows above, 21 Oct 2026, counts updated 8 Nov 2026; 29 integrations, 15 counted, 10 not counted, 4 not settled. Cells (a) to (e) not recorded in the counted rows: 0 of 75 (hermes-jev-mana's no-key cell read 14 Oct 2026; ohmyjev's pi and omp row added 21 Oct 2026; pi-jev-router's five cells read 8 Nov 2026)
- Who publishes it? 1 of 29 comes from the host vendor (OpenClaw's plugin). The other 28 are community projects (GitHub repositories and npm packages). 0 of 29 come from TypeSafe.
- Which mode does it start in? jev-approvals is off after install (Hermes core default
manual; setapprovals.mode: smart); jev-curator starts in observe and blocks nothing until you setguardorapply; jev-judge starts in shadow (log-only); typesafe-skill-router starts off; hermes-jev-mana is inert after the documented install until you setadmit_unlabeled: true. rh-guard starts in enforce, so a default install can block. jev-effort-router is off until you enable it in Hermes, and with Hermes defaults (model"") it is inert; jev-gateway routes by default once a key is set; jev-triage is on at startup; the jev-for-all launcher acts only when you runstart; ohmyjev's pi and omp adapters deny flagged calls only with lockdown on and shadow off (both off by default; read 8 Nov 2026). - What happens with no key? Unset the key and try a risky call. jev-kit, bloudhood and jev-approvals ask you; jev-curator (guard mode) blocks the background deletes and patches it guards; jev-gateway does not start; rh-guard, opencode-tool-gate, osuki and jev-judge (enforce mode) let it through or fall back; jev-effort-router leaves the turn on your configured model; jev-triage passes the message to the main model; the jev-for-all launcher uses its default tier; pi-jev-router sends each request to its standard model; hermes-jev-mana runs the original Hermes branch (judgement); ohmyjev's pi and omp adapters let the call through and log "no Jev key" (3 + 1 + 1 + 4 + 4 + 1 + 1 = 15). For jev-approvals, also make Jev time out: Hermes then asks its main agent model, which can approve (with no main model configured, it asks you). Test procedure: testing without a key.
- Which tools never reach Jev? Read the matcher. bloudhood covers Bash, apply_patch and
mcp__only; rh-guard's Codex matcher has no MCP tools; jev-approvals sees only commands Hermes flags as dangerous; jev-curator sees only background skill deletes and patches; ohmyjev's pi and omp adapters send only Bash, write/edit and exfiltration checks, skip read-only commands and never send other tools. - Is the catalog commit the one you read? Three Hermes catalog entries install a commit other than the repository's HEAD, and jev-curator is in a different repository from the one recorded on 5 Oct (pins below).
- What leaves your machine? See the next section.
What leaves your machine
- Can send file contents
opencode-tool-gate (tool arguments up to 6,000 characters, so content being written), jev-judge (up to 12 arguments, each redacted and cut to 600 characters), jev-review (task, diff, full file contents and repository context, no cap or redaction) and osuki (tool results up to 900 characters, regex redaction only).
4 integrations - Defers secret-like input instead of sending it
jev-kit and bloudhood do not send requests that look like they contain secrets; they hand them to your own prompt instead. bloudhood's endpoint and model are set by you in
settings.json(HTTPS only); no TypeSafe URL is hard-wired.2 integrations
Route data terms are on data privacy.
Hermes Agent catalog: which commit gets installed
The Hermes Agent plugin catalog says: "Every entry installs exactly the commit above" (R9-S64). The rows on this page cite the catalog-installed commit. At least 15 catalog entries mention Jev; 9 were read in source. On 10 Oct 2026 the catalog at main dce1e9b (10 Oct, 03:23 UTC) has 13 jev entries, 12 of them at v2026.9.24; the new one is jev-cron-gate (R12-S210, R12-S211). Documented catalog pages read 6 Oct 2026, 05:17 UTC; HEADs by git ls-remote, 05:14 UTC
| Catalog entry | Catalog installs | Repository HEAD | Note |
|---|---|---|---|
jev (ourines/hermes-jev) | 28a4ea3 (v0.1.2) | cdf59e4 (v0.2.0) | v0.2.0 is not in the catalog; it adds model routing, off by default |
typesafe-skill-router | e6cdac2 | 94fe114 | Page cites the catalog commit |
jev-approvals | 28be98a | 530fdb0 | The two files read are identical at both commits. Read with Hermes core v2026.9.24 (f97608f) on 7 Oct |
jev-curator | 4e8626c | Not compared | Correction: its own repository, anpicasso/hermes-jev-curator. The 5 Oct research put it in the jev-approvals repository |
jev-effort-router | e16f2162 at v2026.9.24; 562dc2a at main dce1e9b | 562dc2a (v0.2.2) | Read 10 Oct at 562dc2a. e16f2162 is one commit behind and changed only audit code; the failure paths are the same (R12-S206) |
jev-judge, jev-typesafe, jev-skill-router-mcp | 5dddbea, b3d29f7, 76ee09e | Same as the catalog commit | DoGMaTiiC's jev-skill-router was read at 5dddbea; its catalog source repository was not verified |
Catalog entries seen by name only, not read: jev-cron-gate (added at main on 10 Oct; per its catalog line a cron pre-run gate that skips an agent run when Jev is confident it is pointless and fails open on every error, Reported), jev-model-router, jev-skill-router; and cards that mention Jev: hermes-slash-router, tool-slimmer, hermes-switchyard, nerve, protean-ordaprompt, local-system-one-hermes, hermes-structured-aux-models.
What is on other pages
- Model routers ("jev codex router": 0xNatoshi/jev-codex-router, kfchow-ai/kfchow-llm-value-router, ourines v0.2.0 routing) pick a model, not whether a tool call runs. Not audited here, except jev-effort-router, the jev-for-all launcher and hermes-jev-mana, which are counted because the host runs Jev's model choice without a person, and Hoppin-Sorter/jev-router, not settled; see model routing.
- Claude Code-only guards stay on Claude Code and MCP guards. jev-kit, jev-review and mcp_jev also target Claude Code; that page points here.
- Library packages (LangChain, Pydantic AI and others) are on framework integrations.
- Every counted row is also on when Jev fails, in the "Coding agents" domain, except ohmyjev's pi and omp adapters: they are part of ohmyjev's row in the "Claude Code / MCP guards" domain (one implementation, read at the same commit with the same five terms; the map still counts 80). You can compare your own design with that map in the failure self-check. There jev-gateway appears as "fallback to the agent's own model" and is one of the 9 fallbacks that can still end in the action (failure map); jev-effort-router, jev-triage, the jev-for-all launcher and hermes-jev-mana are 4 of the 8 routing fallbacks and are not counted as the action going ahead. Across that map, 19 of 80 implementations let the action go ahead on a Jev error (10 fail-open and 9 fallbacks that can still end in the action, jev-approvals and jev-gateway among them; 2 of the fallbacks are real-time demos that keep their last action), recomputed 18 Oct 2026 (17 of 77 on 12 Oct).
- The 18 read by 8 Oct, the 2 counted rows added 10 Oct and the 3 counted rows added 11 Oct are listed in the products ledger, and ohmyjev is in its 18 Oct additions; the 4 not settled are not.
What was not verified
- No agent, hook, plugin, launcher, proxy or MCP server was installed or run, and no Jev call was made. npm tarballs (
@openclaw/typesafe2026.9.8 and osuki 0.2.7 on 6 Oct; osuki 0.2.7 and 0.2.8 on 7 Oct; jev-gateway 0.5.1 on 10 Oct; jev-for-all 0.7.1 on 11 Oct, read in memory) were downloaded and read, not executed, then deleted. Every cell is read from code; behaviour at run time may differ. - What Codex and Cursor do after a hook decision, and what OpenClaw core does with
decision_evaluate. OpenClaw'sbefore_dispatchpath was read on 11 Oct ataa6008a(2026.9.8); at 2026.10.1 (9cf7fc2) it was read only up to L650, so the return statement there was not read. - Hermes core (
v2026.9.24): read on 11 Oct, the defaultmodelis"", so no main provider is set by default (R13-S217), and_BILLING_PATTERNSis inagent/error_classifier.pyL103–118 (R13-S218). Whether jev-approvals' no-key message matches a billing pattern was not checked.jev_policy.pywas not re-read on 8 Oct. - jev-curator: how long a cached judgment from an earlier run stays valid (it is reused without a new call,
engine.pyL80–83) and whether it still covers an operation after a Jev error. - Whether the osuki 0.2.7 and 0.2.8 tarballs match their tag files file by file. osuki's repository address was not recorded, so its row links npm.
- rh-guard's redaction. How OpenCode maps
writeandpatchto permission keys (it does not change the outcome, because"*": "allow"covers every key). - The contents of all test files (file names only).
- Commit dates for 4 MCP repositories (jev-review, burnigtm/jev-mcp, minhgv/jev-mcp, mcp_jev).
- jev-effort-router: whether its
llm_requestmiddleware needs a capability grant in Hermes (the enable step itself was read on 11 Oct). - jev-gateway: how Codex and OpenCode execute a forced or built tool call, and whether their own approval policy applies.
- hermes-jev-mana: the
jevCLI that holds the key (the no-key cell rests onjevkit/client.py, read 14 Oct 2026) and the base Hermes commite97923c3(finish_text_response,run_tool_round, and whetheragent/jev_sourceexists there). - jev-for-all: the Claude Code hook (its source is not in the npm tarball) and OpenCode's handling of tools removed from the list.
- Hoppin-Sorter/jev-router: how the Claude Code engine applies the
turn.stepmodel rewrite (engine source not public). - ohmyjev on pi and omp: what the host does after a non-blocking return rests on the hosts' extension docs (a handler "may block") and ohmyjev's README ("doesn't ask", Reported); pi's and omp's own code after the return was not read.
- auto-mode: how the Claude Code mod host handles the
tool.checkanswer, and the internals of the stages before Jev (7 at 4.0.0). - jcode: callers outside memory, browser and voice (TUI, provider-core, overnight-core).
- acmelabs-15/jev-gate (30 Oct 2026): Claude Code's own handling of the hook's
permissionDecision(only the hooks reference was read); whether anyone uses it; files outside the five entry points and thejev.ts,config.tsandstate.tsexcerpts read. - The 10 lines from the 21–30 Oct scans rest on a README and at most one listing each; their behaviour was not read in source.
- auto-mode 3.1.5 (30 Oct 2026): compared with 3.0.0 in a few files only, not re-read. auto-mode 3.1.6 (released 31 Oct 2026, 09:46 UTC): contents not read beyond the six cited files, which match 4.0.0 (compared 9 Nov 2026). ohmyjev 0.13.2: re-read at
d307c39at the gate lines, 2 Nov 2026; its other pi and omp handlers (routing, screens, done-check) were not re-read. - 8 Nov 2026: ohmyjev 0.14.0 re-read at
fbc4ea4at the gate lines;adapters/codex.tswas not read;bin/ohmyjev.mjs(an install and key CLI, not a gate path) was read only for its usage and key check; the routing, screens, done-check and compaction handlers were not re-read. @caph42/pi-jev-router: read once at the pin with pi's host path; the configured fallback models, pi's code after the routed request is sent and whether anyone uses it were not checked; nothing was installed or run. auto-mode 4.0.0 (8 Nov 2026): not read that day (read 9 Nov 2026). - 9 Nov 2026: auto-mode read at 4.0.0 (
7b246a7) in the six cited files and the npm package's entry points; the mod host's handling of thetool.checkanswer and the internals of the 7 skip stages were not read; the review judge insrc/judge/*was read only at its call site (classify-action.tsL524–545, after a deny); nothing was installed or run. - The 31 lines triaged on 11 Oct rest on READMEs and package descriptions only; no source was read. The 13 lines from the 19–20 Oct scans rest on one file each at a pin.
- The source of jev-cron-gate and of the 6 candidate gates still read from README only; kalowery/jev-router returned HTTP 404 on 10 Oct 2026, 05:13 UTC.
- What jev-effort-router, jev-gateway, jev-triage, the jev-for-all launcher, hermes-jev-mana and ohmyjev's pi and omp adapters send to Jev (not recorded).
- Whether any of these is used in production: none verified.
Leads found but not read (names only, not audited)
- GitHub search hits: qkal/Canny, onlyjq04/jev-agent-hooks, Yaxun-Yang/codex-effort-with-jev, leonaaardob/fast-dev-compaction, 455-dIAO/windows-save-token-jev-setup, Nanako0129/stingray, smixs/code-quality, 0xNatoshi/jev-codex-router.
- Daily scan: bragamat/jevkit, fallen-blossom/flower-control, Steven2007yhq/Reinforced-Computer-use-in-Codex, justmytwospence/opencode-lean-context and opencode-auto-effort,
omo-jev-plugin,omp-typesafe,@geminixiang/pi-jev, n-r-w/classifier-mcp, Rikinshah787/dotpals. - Seen in the 20–21 Oct scans, names only (not classed): Civalski/claujev, acmelabs-15/jev-gate, GilBar2/jev-orchestration, cathy-kim/jev-workflow, felinae98/cmd-mod-jev-guard (Command Code), cognixws/picode-decisions-jev, two DeepSeek Harness plugins (Han-1413141/dsh-dev-thinkout, Tintac-CN/dsh-aidlc) and npm
@bismawy/pi-jev-eye0.3.0. - abaljeu/jev-mcp describes itself as "Patched jev-mcp for thejevai.com … Use with Cursor Cloud Agents MCP" Reported. thejevai.com is not a TypeSafe domain. How to tell official endpoints from others: official vs reseller.
- The 3 Hermes catalog
jev-*plugins still seen by name only, listed in the pins section. The 35 leads triaged on 10 Oct and the 31 lines triaged on 11 Oct are in lead triage.
Sources and check times (6 to 11 Oct, 19 to 21 Oct, 30 Oct, 2 Nov and 8 Nov 2026, UTC)
Per-row sources: pinned commits and file paths
- jev-kit: R9-S72, R9-S73, R9-S75, R9-S85, R9-S86. keiffff/jev-kit at 3d74b05f85c1344ce4ae4b3f07b2eaaa6f957b07 (read 05:20).
packages/hook-adapters/src/index.tsL24–31, L73–90;packages/agent-review/src/index.tsL112–123, L134–140, L151–158, L162–164, L238–283, L295–308;packages/decision-contract/src/index.tsL142, L153–156, L160, L214, L217, L223;packages/cli/src/cli.tsL76–81, L189–192;examples/codex-config.toml,examples/claude-settings.json. SDK@typesafe-ai/sdk0.6.0dist/index.mjsdocstring (R9-S75). - bloudhood/codex-jev-approval-hook: R9-S81. at b6242c9802724550b54eb346d1df40adf997253d.
hook.ps1L13–24, L318–322, L334, L373, L451, L458–460, L581–595, L597, L598–601. Spot-checked against the raw file at the pin (05:22). - 24601/rh-guard: R9-S82; 7 Oct: R10-S117, R10-S118. at c2e682ef0e838cc0538a7c2062e8a96d262dbd14.
src/lib/risk/score.tsL91–106, L110–116, L136–146, L171–181, L239–250;src/lib/risk/jev.tsL75–78, L240, L245–247, L265–279, L335–338;hooks.tsL161–167;hosts.tsL194–195;hooks/run.tsL41–46. Spot-checked against the raw file at the pin (05:22). 7 Oct (read at the pin, 05:20):src/lib/risk/kinds.tsL36, L39–61 (block thresholds), L176–189 (default mode enforce);score.tsL98–109, L125–146, L171–199, L236–241. Same commit, still HEAD on 7 Oct. 8 Oct (R11-S106; read at the pin 05:14 UTC):score.tsL98–127, L42–48, L72–76;kinds.tsL36;jev.tsL78, L265–292. - justmytwospence/opencode-tool-gate: R9-S74, R9-S85. at bdf1ef8d132a16db73814e94b23f8396b3976cac (read 05:20).
src/index.tsL42–43, L240–265, L288, L313, L328–349, L352–380;src/jev.tsL33–36, L47, L51–60;src/judge.tsL147–182. 7 Oct: R10-S114, R10-S115, R10-S116. OpenCodev1.18.35(latest release, published 6 Oct 2026, 20:18 UTC) at anomalyco/opencode 53d1eabb61e21162157817bf677da0a4ad3332e3 (sst/opencoderesolves to the same repository):packages/opencode/src/agent/agent.tsL119–151 (default"*": "allow"; user config merged last);packages/opencode/src/permission/index.tsL28–38 (last matching rule wins).agent.tsspot-checked against the raw file at the pin. @osuki-dev/opencode-osuki-agent0.2.7: R9-S80. npm 0.2.7 (4 Oct 2026, 10:52 UTC); tagv0.2.7=c96ce03c1c7dfc5c6fdf9d966691549101c68080. Tarballdist/index.jsL59–68, L335–337, L343–394, L415–420, L969–1036, L1668–1737, L1751, L1764, L1829–1832. Read once at the pin. 7 Oct: R10-S112, R10-S113. npm 0.2.8 (5 Oct 2026, 11:40:53 UTC; npm sha1dc959c68bd27836f9ece153b023e947efe948404; tagv0.2.8=0ae9e884a97a5539fef3f0772318a628c1f75e00), latest on npm on 7 Oct.dist/index.jsdiff against 0.2.7: the Jev call, timeout and fallback lines (L63, L343–394, L969–972) are identical; hook lines now L1677–1746, L1760, L1838–1841 (0.2.7 L1668–1737, L1751, L1829–1832). Non-failure changes: read-only roles always go to the explore agent (L993–994); newrequireDispatchchecks inosuki_route. Tarball not matched to the tag file by file.- DoGMaTiiC/hermes-jev
jev-judge: R9-S63, R9-S64. at 5dddbeaac5be08281e6c3b372eeaba3639971f0a.plugins/jev-judge/__init__.pyL43–58, L64–73, L77–91, L93–96;jev.pyL44, L52, L281–283, L349–388, L397–401, L468–473, L480–482;gate.pyL277–284, L345–357, L362–368;plugin.yaml.plugins/jev-skill-routerL194–200, L224–226 (read once at the pin, partial). jev-judge Spot-checked against the raw file at the pin (05:23). @openclaw/typesafe2026.9.8: R9-S69. openclaw/openclaw at aa6008ad198ef99c43f9d89dbd01694708712974 (npm provenance commit).extensions/typesafe/src/decisions.tsL28–30, L80–82; tarballdist/index.jsL96–104,dist/.setup/errors-BdIwQubl.mjsL26–30, L45,client-CRLm45hB.mjsL244–256. Spot-checked against the raw file at the pin (05:23).extensions/typesafe/package.jsonat that commit says 2026.9.7; the published version is 2026.9.8.- ourines/hermes-jev: R9-S66, R9-S60. at 28a4ea390f80d97fcdc8c20d6b630dfe2de645de.
service.pyL45, L73–80 (HEAD: L95–101, L149–177);client.pyL132–148. Read once at the pin. - ajensenwaud/hermes-jev-plugin: R9-S67. at b3d29f71770a3447ad0b3db00658f64a8edc7dd3.
client.pyL22, L31–34, L132–158;tools.pyL43–44, L59–64. Read once at the pin. - ydmw74/jev-skill-router: R9-S68. at 76ee09e049dbc10c76f98cfefc6ca16b42cfad38.
src/jev_skill_router/mcp_server.pyL150–158, L167, L169–173, L346–349. Read once at the pin. - NiazMorshed2007/jev-review: R9-S76. at 57690af54ef7d862c2483342c1e61c14dffcf727.
src/mcp/server.tsL43–58;src/evaluation/input.tsL5–37;src/config/environment.tsL3–8;src/jev/client.tsL42, L71–81, L92–118. Read once at the pin. - burnigtm/jev-mcp: R9-S77. at 9448f6120015f2f6c6dad7137c7f234545209918.
src/cli.tsL37–39;src/typesafe.tsL124–128;src/config.tsL214–222;src/errors.tsL53. Read once at the pin. - minhgv/jev-mcp: R9-S78. at 1a1f0a5b599e3c9f00a126bb247536a624aa265e.
src/ci-shadow.tsL6–31;src/typesafe.tsL47–51;src/server.tsL236–244. Read once at the pin. - pedroknigge/mcp_jev: R9-S79. at 32377f8e4d5a63a76d4efbe9abf0b2fbf3f98a0d.
src/packs/command-risk.tsL7–8;src/typesafe.tsL33–35;src/config.tsL9, L48;src/server.tsL21–27. Read once at the pin. - DECRUX9812/typesafe-skill-router: R9-S65. at e6cdac26f9ed588b4a94b8a2f7f9f026e1b9faf3.
__init__.pyL49, L55, L180–190, L242–247, L284–289;typesafe_router/client.pyL292–306. Read once at the pin. - wellkilo/codex-jev-preflight: R9-S83. at 633ddefe6596aa5d7c6b00767c08d07edea61096.
jev_user_prompt_hook.pyL25, L175–196. Read once at the pin (partial). - anpicasso/hermes-jev-approvals (row): R9-S61, R9-S60, R10-S41. Plugin at 28be98a19539b29768398fb400f4c62f41337c26:
plugin/__init__.pyL100–105, L336–341, L495, L520–527, L543, L851, L911–931, L966–974 (read 6 Oct);plugin/jev_policy.pyL13, L54–59 (confidence 0.55; read 7 Oct, 05:25). Hermes Agent core (7 Oct, read 05:18–05:20): R10-S101 releasev2026.9.24= NousResearch/hermes-agent f97608f178d1ffeca59860195ab7da295f7c8e5f; R10-S103tools/approval.pyL774–791, L808–815, L884–887, L918–919, L1178–1192, L1208–1209; R10-S102tools/approval_smart.pyL80–131; R10-S104agent/auxiliary_client.pyL6097, L6222–6230 (30 s default); R10-S105tools/approval_context.pyL228–241 (modemanual, approval wait 300 s). Spot-checked against the raw files at the pin:approval.py,approval_smart.pyand the tag SHA. 8 Oct (read at the pin 05:15–05:16 UTC): R11-S110tools/approval_smart.pyL80–131 (L109–112call_llm(task="approval", …)); R11-S111agent/auxiliary_client.pyL3199–3275, L3289, L6032–6033, L7329–7336, L7610–7701 (L7672–7675_try_main_agent_model_fallback), L7989–8019; R11-S112 pluginplugin/__init__.pyL100–105, L320–341, L481–543 (L531–536 error wrapping), L546–552, L966–974. Spot-checked against the raw files at the pin:auxiliary_client.pyL7625–7675 (05:21 UTC). - anpicasso/hermes-jev-curator (row): R9-S62; 7 Oct: R10-S106 to R10-S111. at 4e8626c7d394a9cbef8594993ec4b208ffb3395f.
plugin/plugin.yamlL24–29 (default observe), L50–53 (25 s);plugin/guard.pyL35, L43, L57–74, L143–160, L376–396;plugin/service.pyL79–104;plugin/debounce.pyL88–96, L107;plugin/engine.pyL80–95, L157–166;plugin/commands.pyL102–104, L120–124. Read once at the pin. 8 Oct (read once at the pin 05:15 UTC): R11-S108plugin/transport.pyL118–122, L128–131, L146–191, withengine.pyL80–83, L113–114 andguard.pyL144–160; R11-S109plugin/graph.pyL21–24, L165–176. - AlphaPerseii3000/jev-effort-router (row, 10 Oct, read at the pin 05:12–05:13 UTC): R12-S200 to R12-S207. at 562dc2a27aa22977e0795809a17fbbbb1819f0ab (v0.2.2).
__init__.pyL68;router.pyL105–137, L140–149, L153–161, L177–192, L233–235, L250–253, L256–277;client.pyL178–179, L193–199, L204–214, L222–228, L237–252, L267–270, L279–285;config.pyL29–34, L90, L124, L147–152;plugin.yamlL33–35, L43–47. Comparee16f2162...562dc2a(R12-S206). Catalog entry atf97608fanddce1e9b(R12-S207). Hermes host atf97608f:hermes_cli/middleware.pyL63–99 (R12-S208),agent/turn_api_request.pyL140–151 (R12-S209). Below-threshold cell Spot-checked against the raw file at the pin (05:19:56 UTC). - jev-gateway (row, 10 Oct, read at the pin 05:14:50 UTC): R12-S257. npm 0.5.1, gitHead
37ff982949f0cb9bb131963f82a6f34f8ba38edd; tarball read in memory, not executed.decide.jsL12–28, L102–110, L122–133, L141–145, L150–157, L161;config.jsL24, L39, L43–44;jev.jsL65–66, L90–101;index.jsL10;launcher.mjsL112–119;app.jsL117–118, L150–153, L205–212. Not spot-checked. - RanyAlbegWein/jev-triage (row, counted 11 Oct; read at the pin 05:17–05:18 UTC): R13-S200 to R13-S207; 10 Oct: R12-S254. at 6360e217e61d469fd4cb4798c1d81dc6b722534f (1.0.3):
index.jsL32–49 (L34, L39, L41, L42–44, L45–47; R13-S200);openclaw.plugin.json(activation onStartup; R13-S202);README.md(Reported; R13-S201). Host openclaw/openclaw at aa6008ad198ef99c43f9d89dbd01694708712974:src/auto-reply/reply/dispatch-from-config.choose-route.tsL558–562, L575–617, L618–642, L657, L663–692 (R13-S204; Spot-checked against the raw file at the pin, 05:28 UTC);src/plugins/hooks.tsL256–276, L514–524, L789–804, L829–840, L842–843 (R13-S205);src/plugins/hook-runner-global.tsL22–30 (R13-S206);src/plugins/hook-types.tsL478–481 (R13-S207); tree R13-S203. Drift:@openclaw/typesafe2026.10.1 provenance 9cf7fc279c26cf3493b75e703aeadac6d0deb5ee (R13-S208);choose-route.tsL618–650 read (R13-S209). - abehrman/hermes-jev-mana (row, counted 11 Oct; read at the pin 05:18–05:20 UTC): R13-S210 to R13-S216; 10 Oct: R12-S255, R12-S256. at 3daf639c148ccc10c570b87b78b6549c244839b4:
patches/jev-decisions.patchread in full, 2,079 lines (L88–93, L106–118, L131–137, L428–429, L436–438, L442–458, L937–941, L1062–1078, L1116–1201, L1357–1397, L1782–1836, L2010–2020; R13-S210; L88–93, L428–429 and L1396–1397 Spot-checked against the raw file at the pin, 05:38 UTC);tools/jev_mana_install.pyL1–15, L334–337, L401–402 (R13-S211);plugin/jev-mana/decision_hooks.pyL61–71, L76–90 (R13-S212);plugin/jev-mana/jevkit/decisions.pyL231–287 (R13-S213);config/jev_config.example.yaml(R13-S214);README.md(Reported; R13-S215). Hermes tree atf97608fhas noagent/jev_source(R13-S216). jev-for-all0.7.1 (row, launcher counted 11 Oct; read at the pin 05:20–05:21 UTC): R13-S222 to R13-S224. npm 0.7.1, gitHead74f62e71ee12ae29ce278c6e1005feec0e78de46; tarball read in memory, not executed. Launcher:bin/pick.tsL55–72;src/models.tsL20–75;spec/decisions.jsonL76–80, L97–101, L119–123 (R13-S224). OpenCode plugin (not settled):index.tsL85–88, L93, L113, L237–400 (R13-S222);src/tools.tsL63–133 (R13-S223). Read once at the pin.- Hoppin-Sorter/jev-router (not settled; read at the pin 05:21 UTC): R13-S225 to R13-S228. at bdb47215570e8e41a7a7b8e4c3d6ce91a12674a2 (0.7.0):
hooks/register.tsxL58, L88, L114–127, L143–177, L349–351, L364–374, L395–447, L484–497 (R13-S226);lib/jev-router.tsL133–139, L152–159, L228–255 (R13-S227);.claude-plugin/plugin.json(R13-S228); tree R13-S225. Read once at the pin. - Hermes Agent core defaults (11 Oct, read at the pin 05:19–05:20 UTC): at f97608f178d1ffeca59860195ab7da295f7c8e5f:
hermes_cli/config_defaults.pyL13, L20–22 (R13-S217; Spot-checked against the raw file at the pin, 05:28 UTC);agent/error_classifier.pyL103–118 (R13-S218);hermes_cli/plugins.pyL1370–1381 (R13-S219);hermes_cli/plugins_cmd.pyL986–999 (R13-S220);plugin-catalog/jev-effort-router.yaml(R13-S221). - Novacon/ohmyjev, pi and omp adapters (row, read at the pin 19 Oct 2026; re-read 21 Oct 2026, 05:11–05:13 UTC): R21-S100 (commits feed: HEAD
78bb82c, 15 Oct 2026), R21-S101 to R21-S107. at d307c399fbc4898200fb2c229c623df9a517f318 (0.13.2; R26-S20 to R26-S24, 2 Nov 2026):adapters/pi.tsL449, L497–560;adapters/omp.tsL413, L522–600;adapters/shared.tsL28, L37, L106;hooks/policy.tsL12, L57–58, L65, L84, L390–428;hooks/ohmyjev.tsL272. at fbc4ea49dc5d3e2bd9c164a5b57563739daa442e (0.14.0; R28-S20 to R28-S23, 8 Nov 2026):adapters/pi.tsL498, L504, L508, L522, L538, L602–606;adapters/omp.tsL524, L531, L535, L563, L590, L678–682;adapters/shared.tsL173–182;hooks/policy.tsL12–13, L33–39, L370, L414, L424–425. Earlier at 78bb82cea03c0fb06fbf8892dc6d1840a7f02c11:adapters/pi.tsL447, L465–547, L585–605;adapters/omp.tsL411, L520–600, L634–725;adapters/shared.tsL26–47, L105–127;hooks/policy.tsL18–26, L63, L323–371;hooks/jev.tsL33–66;package.jsonL40–48; README L110, L176. Host docs: pipackages/coding-agent/docs/extensions.mdL218 at16f5f0a(R21-S109, R21-S111); oh-my-pidocs/extensions.mdL548 atd5cd750(R21-S110, R21-S112). Spot-checked against the raw files at the pin. - zgeoff/auto-mode (not settled; 21 Oct 2026, 05:13–05:16 UTC): R21-S115 to R21-S138. Read at 458c41c (2.0.2); cited files re-read at c0b9378 (3.0.0).
mods/auto-mode/hooks/register.tsL110–211;src/config/config.tsL69, L76–77;mods/auto-mode/contract/cli-limits.tsL2;src/model/classify-with-jev.tsL48–51, L111–122;mods/auto-mode/hooks/parse-decision.tsL5–35;src/classify-action.tsL228–235. Read once at the pin. 9 Nov 2026: R29-S200 to R29-S207, R29-S17. Read at 7b246a7 (4.0.0, tag auto-mode-v4.0.0; the six cited files are identical to 3.1.6, 382b7a9):mods/auto-mode/hooks/register.tsL118–120, L198;src/config/presets.tsL15;src/config/config.tsL69, L76–77;mods/auto-mode/contract/cli-limits.tsL2;src/model/classify-with-jev.tsL51–54;mods/auto-mode/hooks/parse-decision.tsL3–33;src/classify-action.tsL234–242, L470–488, L524–545. Read once at the pin. - 1jehuang/jcode (not settled; 21 Oct 2026, 05:12–05:13 UTC): R21-S230 (PR #1890), R21-S231 (source at f5c49226953ec39c1ecfeb4c67317869c0d03add;
memory.rsL1181, L1190;browser_jev.rsL2;voice_intent.rsL193–199). Read once at the pin. - Lead triage, 19–20 Oct (13 lines): R21-S200 to R21-S227, one file each at the pins listed, 21 Oct 2026, 05:12–05:20 UTC. Kilo Code searches: R21-S228, R21-S229. Class counts by script.
- acmelabs-15/jev-gate (not settled; read at the pin 30 Oct 2026, 05:21–05:28 UTC): R24-S228 (commits feed: HEAD
b3ae111, 20 Oct 2026, 13:38 UTC), R24-S229 (tree at the pin), R24-S230 (hooks/hooks.json), R24-S231 (scripts/gate/hook.ts), R24-S232 (scripts/gate/jev.ts), R24-S233 (scripts/gate/config.tsand the other entry points), R24-S43 (scripts/report/report.ts, 278 lines, andscripts/gate/session-start.ts, 53 lines: no Jev call), R24-S234 (Claude Code hooks reference; host source not read). at b3ae111c09e205bf132ff7f7f3a6c2d2c8558298. Spot-checked against the raw files at the pin (error or timeout, below-threshold and bypass cells). - Caph-dev/pi-jev-router (counted; read at the pin 8 Nov 2026, by 05:28 UTC): R28-S200 to R28-S203, R28-S205 to R28-S208, R28-S210, R28-S48 (npm registry and tarball of
@caph42/pi-jev-router2.0.0, gitHead75ad5b7; at 75ad5b71940624ba29baf39ad0e7c51319db598e:extensions/jev-router.tsL115–135, L196–201, L288, L431–493, L820, L870, L915–929, L992–1113, L1149–1182; pi 1.5.0 at f9a26365568127461c47a07a37dcb42c140a3ac5:packages/coding-agent/src/core/model-runtime.tsL677, L829–830, L972–980, L1011–1041;agent-session.tsL830–858). - Lead triage, 21–30 Oct (10 lines): R24-S235 to R24-S245, README and at most one listing at the pins or versions listed, 30 Oct 2026, 05:23 UTC. Reported.
- Version checks, 30 Oct 2026, 05:20 UTC: R24-S35 (npm registry,
@openclaw/typesafe2026.10.9), R24-S37 (auto-mode 3.0.0 and 3.1.5 package files compared), R24-S38 (ohmyjev 0.12.0 and 0.13.2 policy files compared, 30 Oct); cells re-read atd307c392 Nov 2026 (R26-S20 to R26-S24). - Version checks, 2 Nov 2026: R26-S20 to R26-S24 (ohmyjev 0.13.2 at
d307c39), R26-S36 (npm registry: auto-mode 3.1.6, released 31 Oct 2026, 09:46 UTC; contents not read). - Version checks and reads, 8 Nov 2026: R28-S20 to R28-S23 (ohmyjev 0.14.0 at
fbc4ea4, npm registry and raw files, 05:19–05:21 UTC); R28-S200 to R28-S210 and R28-S48 (@caph42/pi-jev-router 2.0.0 at75ad5b7, pi 1.5.0 atf9a2636); R28-S47 (npm registry: auto-mode 4.0.0, published 8 Nov 2026, 01:03 UTC; contents not read). - Version checks and reads, 9 Nov 2026: R29-S200 (npm registry: auto-mode 3.0.0, 3.1.6 and 4.0.0), R29-S201 to R29-S203 (npm package files for 4.0.0, 3.1.6 and 3.0.0, read only), R29-S204 (tags feed), R29-S205 to R29-S207 (source at tags auto-mode-v4.0.0
7b246a7, auto-mode-v3.1.6382b7a9and auto-mode-v3.0.0c0b9378), R29-S17 (src/classify-action.tsat7b246a7, L234–242). - Pinutss jev-agent-router, jev-mcp-router, jev-memory-selector (note): R12-S212 to R12-S218; READMEs at
8cb6d67,6a7b3ba,87bd9be. Reported. - Hermes catalog at
main: R12-S210 (dce1e9b37581dd62e480a9064dc04a709c2940d3, 10 Oct 2026, 03:23:21 UTC), R12-S211 (plugin-catalog/jev-cron-gate.yaml). Catalog atf97608f: R11-S301. - Lead triage (35 leads): R12-S220 to R12-S253, READMEs and package descriptions, 10 Oct 2026, 05:13 UTC. Class counts by script.
- Lead triage, 31 lines (lines 36–66): R13-S229 to R13-S248 (the 20 items from 10 Oct 2026, READMEs at the HEADs listed,
git ls-remote05:21 UTC) and R13-S249 to R13-S259 (the 11 items from 11 Oct 2026, 05:22–05:23 UTC; R13-S258 no HEAD; R13-S259 npm registry). Reported.
- R13-S26: npm registry search, 11 Oct 2026, 05:21 UTC:
@openclaw/typesafe2026.10.1 published 10 Oct 2026, 22:13 UTC (drift line). - R12-S11: search suggestions from Google, Bing and DuckDuckGo, 10 Oct 2026, 05:17:33–05:17:36 UTC. A demand signal, not a volume.
- R12-S02: GitHub repository searches for the 8–10 Oct scan window, 10 Oct 2026, 05:17:31 UTC.
- R12-S48, R12-S49: npm, PyPI, Maven and GitHub release rechecks for the drift line, 10 Oct 2026, 05:18:47 UTC.
- R9-S11: search suggestions from Google (
suggestqueries,client=firefox), Bing (osjson) and DuckDuckGo (ac), 05:19:23–05:19:30. A demand signal, not a volume. - R9-S38: TypeSafe's GitHub organisation repository list (12 repositories), 05:15:34.
- R9-S60:
git ls-remoteHEAD for 6 Hermes plugin repositories, 05:14:50. - R9-S64: Hermes Agent plugin catalog pages (7 entries) and the rendered index, 05:17:40.
- R9-S84: GitHub repository searches "jev codex hook" (26 results), "typesafe cursor hooks" (1) and "jev codex router" (96), 05:17:45.