Research library
Research reports for this catalogue
Thirty-nine dated reports document evidence, decisions and limits. The latest one-URL evidence report separates Google generative impressions, Bing-supported citations, analytics visits and single answer sightings from page checks. It adds no score or fifth card. Four candidates remain conditional and untested, and two earlier candidates remain historical nonrecommendations.
Research collection
Use List, Table, or Calendar to view the same thirty-nine reports. Each report has one permanent record in this library.
Read the report limits before using a result
One owned URL: appearance evidence
Google generative Search impressions, Bing-supported citations, first-party referral visits and manual answer sightings have different definitions. Record them separately with dates, canonical mapping and reviewer; no property data or causal test was available. Read the report.
Two jobs: local endpoint and connected-action stop
The 30 October comparison sets a local endpoint, missing input, named reviewer and action stop for an approved-article social draft and a landing-page experiment preflight. Buffer and Statsig illustrate later connected paths, not observed benefits or configured permissions. Read the report.
Two local social-draft routes
On 29 October, a maintained template and manually source-guided route each produced LinkedIn and four X drafts from the published A/B-test report. Both needed numeric caveat corrections. The author reviewed the one artifact unblinded; no comparative fidelity or effort advantage was measured. Read the full report.
A/B-test preflight: worksheet and analyst first
The 28 October source has useful hypothesis and metric prompts. Its rounded sample table is plausible under a conventional approximation, not a page-specific power plan. Implementation, monitoring, inference and release need separate owners. No live test or winner exists. Read the full report.
Social Content: local drafts only
The 27 October source read supports two local text formats from an approved article, not account actions or measured reach. A maintained template and reviewed prompt come first; the exact package, host access and advantage over that route are unknown. Read the full report.
Active-card source-link check
The 24 October connector resolved four original selected files. Three mutable links match their labels; the primary emails citation now uses the inspected v2.11.37 tag. Broker cache miss and DNS failure do not establish a broken tag or a fresh browser response. No package or host was tested. Read the link check.
Optional competitor-content source
The 23 October audit reads the separate Competitor Content Analysis selected file. A sampled URL inventory can inform a brief; estimated traffic, gaps and impact order need dated evidence, paired coverage and content-owner review. No skill or target site was tested. Read the full report.
Competitor-to-content handoff boundary
Run 32 permits only owner-confirmed, dated site observations to enter a separate content brief. Discovery rank is not coverage or priority; customer need and capacity require their own evidence. No two-skill route was tested. Read Run 32.
Copywriting pre-install evidence ledger
Run 31 separates exact payload, chosen-host discovery, effective access, claim review and publication approval. It reuses earlier evidence without a new source or host test. Read Run 31.
Source inspection
The first report reads original repository pages and source files. It distinguishes maintainer statements from this pool’s observations.
Comparison
The second report compares the same five records by output, stated access, data surface, and required review. It adds no source inspection, installation, or accuracy testing.
Pinned provenance & freshness
Run 3 captured a dated Copywriting selected-file commit and release record. For the four Superamped entries, selected-file commit/date and tag findings remained unresolved.
Safe evaluation boundary
Run 4 requires a stop before an install recommendation for the four Superamped entries and human approval before an external action.
Provenance recheck
Run 5 did not obtain a selected-file SHA/date or tag finding for any of the four named Superamped paths. The release page says the repository has no releases, but it has no direct selected-file coverage. The stop remains.
Lifecycle/email candidate review
Run 6 keeps Core Haines emails only as a conditional, draft-only candidate. It requires an owner-approved non-personal brief, stops on unapproved local context or linked tools, and requires separate human approval before any email-system action. RampStack email-sequences is excluded.
Current-source recheck
Run 7 found no evidenced substantive claim change, but could not confirm a current-day directory listing for any of the six selected packages. Cached source pages do not verify current package contents, installability, or compatibility.
Monthly briefing job
Run 8 proposes an offline, human-reviewed brief from approved aggregate channel exports. Community posts are qualitative signals, not prevalence or outcome evidence. No reporting package was sought or inspected, and the catalogue stayed at six at that date.
Monthly reporting shortlist
Run 9 inspected Anthropic and Skill Me reporting files. Anthropic allows pasted metrics but omits mandatory audit and human-release steps; its wider plugin supports connectors. Skill Me explicitly instructs sending. Neither qualified; no package was run or then-six-entry card rechecked.
Reporting search closed for now
Run 10 inspected ClawFu and Digital Marketing Pro. ClawFu directs dependency installation and a script; Digital Marketing Pro has real source-of-truth and teammate checks, but not full claim trace, metric reconciliation or authorized-owner sign-off. Neither earns a seventh card or test. This was bounded, not exhaustive.
Two-channel launch brief
Run 11 defers a package search: practitioner discussions show brief-making, but not unmet need for a portable skill over simpler alternatives. It defines a low-access, owner-reviewed first use and a prelaunch-QA research question. No package was sought or run.
Prelaunch setup QA and UTM governance
Run 12 keeps these as separate jobs with a narrow tracking-string check at their interface. It defers a package search: no evidence shows a portable skill beats a spreadsheet checklist and human review. A local discrepancy is not a live defect or launch clearance.
Competitor messaging-change brief
Run 13 defers a package search. Compare authorized, dated captures for one known rival, preserve exact source lines and context, and have a named owner judge materiality. A saved diff, versioned sheet, reviewed prompt or page monitor may suffice. No package or monitor was tested.
Landing-page copy route choice
Run 14 recommended a maintained brief/template and reviewed prompt for one approved local draft. Copywriting source then displayed 2.1.0, but current selected-file SHA/date, complete directory inventory, numeric conversion assertions, install and output quality remain unverified. A third-party skill is a conditional later evaluation, not the default.
Route-first six-entry reconciliation
Run 15 keeps non-skill routes available, adds a six-row input/read/output/review/status matrix, and corrects the emails badge to a mutable 2.1.1 source observation. No then-six-package re-audit, install or outcome test. Read Run 15.
Changed-source check
Run 16 confirms file-attributable 2.1.0 changes for Copywriting and emails, not latest selected-file pins. The 2.1.1 emails SHA/date and full diff remain unknown; cross-skill reference and tool links need separate review. Read Run 16.
Ad Campaign Analyzer decision boundary
Run 17 removes the ad package from active choice. Its selected source instructs immediate stop/scale action from unsupported profitability and significance rules. Use a governed worksheet for descriptive export triage; a named paid-media owner validates and separately authorizes any action. SHA/date, tag and current inventory remain open. Read Run 17.
Search Page Audit score boundary
Run 18 removes the scored audit from active consideration. Its mandatory X/38, ratings and fixed impact order do not validate Google Search quality, AI citation or ranking gain. Use an owner-maintained, unscored checklist for one authorized public URL. SHA/date and current inventory remain open. Read Run 18.
Content Strategy priority boundary
Run 19 keeps the card as a conditional draft scaffold. Its weighted X/10 priorities, buyer-stage modifiers and sampled gaps are not validated demand or return. For one month use a versioned brief/template first; the selected-file pin and current inventory remain open. Read Run 19.
Competitor Discovery ranking boundary
Run 20 retains the card for a sampled, human-confirmed starting set, not a market map. The selected source normally excludes tangential results and ranks by query frequency, which does not verify threat or action priority. Use a versioned owner sheet first, with dated claim evidence and identity checks. Selected-file pin and current inventory remain open. Read Run 20.
Welcome-email draft boundary
Run 21 retains the conditional emails card only for our proposed five-email post-signup welcome draft. Owner template first; timing, exits and metrics are hypotheses, not configured rules or targets. A named owner approves proof, terms and the exact CTA URL; ESP action needs separate approval. Mutable 2.1.3 was observed 11 October, not pinned. Read Run 21.
Copywriting claim/proof boundary
Run 22 retained Copywriting as a conditional draft scaffold for one approved local landing page, with an owner template first. Its numeric efficacy claims and referenced 3× example lack a checkable underlying study in inspected passages. AI-tells is optional style review, not validated detection. The pool-proposed claim register and named review hold unsupported proof and CTA links before a separate publication decision. Read Run 22.
Cross-card decision consistency
Run 23 retains four conditional active cards and two historical nonrecommendations. It corrects the standalone answer to name both removed skills and clarifies that five welcome emails are the pool’s proposed subset, not a source-required format. This compares saved evidence and public surfaces only; no source was freshly reopened or route tested. Read Run 23.
Email linked-reference boundary
Run 24 reads the three in-skill references, cross-skill AI-tells page, registry, setup and six directly linked implementation guides. The guides describe credentials, recipient data and send routes, not required draft steps. Their inclusion in one emails install and any host/provider behavior remain unverified. Read Run 24.
Single-skill install documentation
Run 25 finds documented named-skill CLI selection and a one-skill ZIP design, not an inspected payload. It maps in-skill resources against cross-skill and tools/ links, and records Copywriting’s mutable 2.2.0 text and source-directed AI-tells check. No archive, CLI-selected files or host behavior were tested. Read Run 25.
Copywriting source-delta boundary
Run 26 compares full 2.1.0 and observed 2.3.1 files, not a complete intermediate 2.2.0 file. Current source output depends on the input: page draft or narrow edit. Source claim-support directions are distinct from the pool’s proof, rights, CTA and publication gates. Latest pin, full inventory, payload and host behavior remain unverified. Read Run 26.
Email release-asset boundary
Run 27 identifies GitHub-reported metadata for the original v2.11.37 emails-source.zip and a tagged 2.1.4 source file. The ZIP and attached manifest were inaccessible; no entry is confirmed present or missing, and no checksum was computed from bytes. The conditional local-draft decision and recipient-policy/ESP stop remain. Read Run 27.
Tagged email selected-file delta
Run 28 compares complete immutable 2.1.0 and 2.1.4 selected files. New reference and local Learnings instructions increase the review and data boundary; mutable-main tool links do not become a required ESP step for a local Markdown draft. The exact 2.1.3→2.1.4 delta and package content remain unknown. Read Run 28.
Copywriting host-documentation boundary
Run 30 separates official Codex and Claude Code local-skill capabilities from the maintainer’s named-skill and Claude-web ZIP claims. It does not establish installed files, reference loading, effective policy or output quality. Use the owner template first and review exact files, authorized context and permissions before any installation. Read Run 30.
Directory signals versus selection evidence
Run 29 treats directory scores, A–Z ordering, install counts, stars and audit badges as source leads, not a common quality scale or installation clearance. One result on each of three sites was read; scoring detail and one safety detail were unavailable on the inspected path. Start from the original source and a smaller authorized route. Read Run 29.
Not tested
No selected skill was installed, run with credentials, security-reviewed, legal/privacy-reviewed, or tested for host compatibility or marketing results.
Run 3 pinned provenance recheck · Run 5 provenance recheck · Run 6 lifecycle/email candidate review · Run 7 source and package-surface recheck · Run 8 monthly briefing job · Run 9 monthly reporting shortlist · Run 10 reporting search · Run 11 launch-brief decision · Run 12 prelaunch QA decision · Run 13 competitor-change decision · Run 14 landing-page copy choice · Run 15 route-first chooser · Run 18 search-score audit · Run 19 Content Strategy boundary · Run 20 Competitor Discovery boundary · Run 21 welcome-draft boundary · Run 22 Copywriting claim/proof boundary · Run 26 Copywriting source delta · Run 27 email release asset · Run 28 email source delta · Skill catalogue · Methods & limits · Review a skill