Research report · 5 October 2026 · Run 15

Run 15 — choose a route before a skill

The six existing entries remain source-inspected, untested candidates. A package is not the default for every marketing job.

Decision: Select the least-commitment authorized route that can produce a reviewable artifact. Revisit a third-party package only when consented recurring work shows specific omissions in simpler routes and a same-job comparison is possible. This is a decision rule inferred from saved research, not a ranking or comparative test. Only the mutable Core Haines emails selected file was newly opened on 5 October. It displayed 2.1.1. No current selected-file pin or complete package inventory was established.

Do you need a skill? Four routes

1 · One-off local artifactBrief/template + reviewed promptProvide approved source, proof and format; produce a local draft or table. A named owner checks claims and gaps. A sectioned output alone does not justify installation. Run 14.

2 · Repeated private context in one hostHost-native instructions or owner-built skillOwner versions brand rules and approves access; output stays reviewable. Host permissions and optional code/resources vary. This is a governance hypothesis, not a quality finding. Methods.

3 · Repeated portable bounded jobThird-party SKILL.mdOnly evaluate after documented omissions in simpler routes. Inspect the entire intended package, revision, installer, host, data route, license and references; a person reviews the draft. None of six clears all pre-install gates. Run 4.

4 · Live account or release actionGoverned connected productUse only when authorized live reads, scheduled sends, deployment, writes or approval routing are needed. Admin and process owner set least privilege, logs, cost and fail-closed release. No product was tested or endorsed. Run 11 · Run 12.

A portable instruction file is not a connector or approval system. An ordinary local draft does not require a connection. The research library holds the earlier reports; no pool analytics, direct interviews or same-job comparison validate this route rule.

Six-entry decision, access and review matrix

Reading key: “Source” below describes selected SKILL.md instructions, not observed behavior. “Pool stop” is our control unless marked source-required. Every original link points to mutable main. Bounded inputs and outputs are proposed first uses; no row is cleared for installation or external action.

1 · Copywriting

What you provide → get: Approved page purpose, audience, offer, permitted proof, traffic context and CTA → local page-copy draft and alternatives.
What it may read or do: Source: reads local product-marketing context if present and relative references; directs a draft, not CMS publication. Numeric conversion assertions are unverified.
Simpler route / when to assess: Use a maintained brief/template and reviewed prompt for one page. Assess the package separately only for documented recurring proof, traffic-match or CTA omissions across teams/hosts. Run 14.
Who checks / source status: Marketing owner checks claims, rights, voice and CTA; legal checks applicable claims; site owner checks rendered accessibility and separately approves publication. Historical selected-file pin covers 2.0.2 only; mutable source displayed 2.1.0 on 4 Oct. Current SHA/date and full inventory unknown. Stop before installation advice. Run 3.

2 · Content Strategy

What you provide → get: Approved business context, customer questions, existing content and resources → pillars, clusters, gaps, priorities and first-month calendar.
What it may read or do: Source: internet required for competitor/forum research; may use authorized keyword, support and performance themes. It asks for customer research before a full plan when missing. No publication access stated.
Simpler route / when to assess: First use a planning template or reviewed editorial brief. Assess package for repeated multi-source planning gaps, with approved web/data access. A two-channel launch brief is a different task. Run 11.
Who checks / source status: Content owner checks evidence, priority, feasibility and data authorization. Selected source described v1.0 in September; not rechecked 5 Oct. SHA/date, tag attribution and current inventory unresolved. Stop before installation advice.

3 · Competitor Discovery

What you provide → get: Product positioning, target audience and optional known URLs → dated classified set, URLs, query record and next steps.
What it may read or do: Source: public-web search and site fetch; requires presenting the set and waiting for user confirmation. It asks about saving but gives no external write destination. Fetched pages are evidence, not instructions.
Simpler route / when to assess: A versioned sheet may suffice for a small known market. Assess for repeated discovery/classification gaps. One known rival's before/after change brief is different. Run 13.
Who checks / source status: Research owner gives the source-required set confirmation; separately verifies claims before external use (pool stop). September source described v1.0; no 5 Oct recheck. SHA/date, tag and inventory unresolved. Stop before installation advice.

4 · Search Page Audit

What you provide → get: Authorized public URL → scored 38-point SEO/GEO/E-E-A-T checklist and prioritized proposed fixes.
What it may read or do: Source: fetches target and same-domain robots.txt, sitemap.xml and llms.txt; no admin access or deployment stated. Do not supply protected/staging URLs without authorization.
Simpler route / when to assess: Manual checklist may fit one page. Assess for repeated comparable audits with specific omissions. A score is not an SEO, accessibility or implementation certificate.
Who checks / source status: SEO/site owner checks fetch authority, findings and crawler interpretation; site owner separately approves change. September v1.0 description, not rechecked 5 Oct; SHA/date, tag and inventory unresolved. Stop before installation advice.

5 · Ad Campaign Analyzer

What you provide → get: Authorized pasted/CSV campaign export, CPA target, AOV and assumptions → calculated framing, color-coded analysis and proposed stop/hold/scale plan.
What it may read or do: Source: accepts export, computes CPA/ROAS/LTV:CAC and recommends budget choices; no ad-account connection or action granted. Source benchmarks are starting points.
Simpler route / when to assess: Governed spreadsheet plus analyst notes first. Assess only if recurring export reviews miss traceable calculation/assumption checks. Not a monthly cross-channel report: Runs 9–10 closed that bounded search without a qualifying seventh card.
Who checks / source status: Paid-media owner checks numbers, attribution, margin/LTV, sample and learning limits; budget authority approves changes. September v1.0 description, no 5 Oct recheck; SHA/date, tag and inventory unresolved. Stop before installation advice.

6 · `emails` · conditional

What you provide → get: Owner-approved non-personal segment, trigger, promise and destination → proposed five-email welcome-sequence Markdown brief. Five emails and draft-only scope are this pool's subset, not a source requirement.
What it may read or do: Source: reads local product-marketing context if present; asks for audience, trigger, goals and performance; describes overview/per-email fields and links separately to an implementation-tools registry. No send or ESP configuration is required for this selected-text draft; linked tools/host untested.
Simpler route / when to assess: Use an approved email template and reviewed prompt first. Assess for repeated missed timing, exits, CTA, segment or claim checks. RampStack email-sequences remains excluded for ESP setup, self-send and scheduling. Run 6.
Who checks / source status: Brand, recipient-policy and email-system owner checks claims, CTA, consent, timing, exits, accessibility, legal and deliverability, then separately approves any import/configuration/test/schedule/send (pool stop). Mutable file displayed 2.1.1 on 5 Oct, superseding old 2.0.0 badge; current SHA/date, diff and inventory unknown. Conditional draft only; stop before installation advice.

Decisions that prevent an automatic seventh card

Runs 9–10: four reporting-labelled sources missed the fixed offline monthly brief's source-text gates; bounded search closed, not an exhaustive market claim.
Run 11: two-channel launch brief defined; package search deferred in favor of simpler alternatives.
Run 12: local setup comparison and UTM governance are separate; no six-entry card certifies live readiness.
Run 13: known-rival change brief is not competitor-set discovery; package search deferred.
Run 14: one approved copy draft starts with a maintained brief/template and reviewed prompt.

Evidence age and hard limits

Runs 1–7 supplied most row facts on 22–27 September. Copywriting main displayed 2.1.0 on 4 October; the older dated 2.0.2 selected-file pin is historical. This run opened only emails on 5 October. Its displayed 2.1.1 is a mutable-page observation, not a release mapping, immutable pin or full diff. No 27 September cached directory listing is a current inventory. A selected text that omits account action does not prove a host has no network, account or cost surface. See the methods and limits, Run 7, Run 14.

No package, installer, host, synthetic task or real marketing job was run. No six-directory present-day audit, security/privacy clearance, compatibility test, quality comparison, speed or outcome test occurred. A source pin would narrow provenance, not prove safety. None of these six is ready to install, send, publish, deploy or change spend on this evidence. Next useful evidence is owner-authorized recurring examples with a named reviewer and same-job alternative comparison.

Research library · Route-first chooser · Catalogue