Marketing Skills Standard (MSS)

MSS is a standard first, a repository second, and a website last. Any proposed addition gets one question: does this make the standard better? If not, it waits — see WATCHLIST.md.

An open evaluation protocol for AI marketing capabilities — not a directory, not a leaderboard engine. The methodology is the product; evaluations are examples of it in action.

This month's success metric

Not "N evaluations published." Instead: can an experienced marketer read one MSS evaluation and immediately understand why Skill A beat Skill B? If yes, the protocol communicates well. If no, that's a protocol problem, not a scale problem — log it in EDGE_CASES.md, don't patch it silently.

Why

Every "awesome list" of AI marketing skills ranks by GitHub stars, which measures popularity, not whether the output is any good. MSS evaluates real outputs against a fixed, versioned rubric with quoted evidence for every deduction — so a comparison is reproducible, not a vibe.

Structure

• REVIEW_STANDARD.md — the six scoring dimensions, weights, anchors, and the constitution (reproducibility, evidence, versioning, independence).
• TEST_BRIEFS.md — one standard brief per capability. Every skill tested under a capability gets the same brief.
• EVALUATIONS/ — individual evaluation files, one per (capability × skill) pair, identified as MSS-<CAPABILITY>-<SKILL>-<STANDARD_VERSION>.
• research/ — discovery-phase artifacts (candidate lists, raw scrape output) kept for auditability, not polished documentation. See research/README.md.

Scope (v1)

Generative marketing capabilities producing text or strategic recommendations: LinkedIn posts, SEO blog writing, positioning, messaging, ICP generation, email, landing pages, Reddit research, GTM strategy.

Execution tools (CLIs, API wrappers, data connectors) are out of scope for v1 — see REVIEW_STANDARD.md for why, and the planned MSS-Exec standard.

Initial results

Skill	Capability	Score	Standard
kostja94/marketing-skills	LinkedIn Post Writing	8.05	MSS v1
mysticaltech/marketingskills	LinkedIn Post Writing	7.50	MSS v1
gooseworks-ai/goose-skills	LinkedIn Post Writing	7.15	MSS v1

Preliminary — see EVALUATIONS/ for the full write-ups with quoted evidence.

Status

v1 — validation in progress.

• Initial LinkedIn Post Writing capability evaluated (3 skills)
• Discrimination check — does the rubric distinguish evidence quality, not just quantity?
• Repeatability check — blind rescore of an existing evaluation
• Independent reviewer check

See TRACKER.md for details.

Testing

pip install pytest
pytest tests/ -v

No application code here — it's a methodology plus written evaluations. So tests check what actually matters for that shape of repo: every evaluation follows the metadata schema and naming convention REVIEW_STANDARD.md defines, referenced briefs and raw-output files exist, and — the one that matters most — each evaluation's stated Overall score is verified to actually be the weighted average of its own six dimension scores, not just a number typed in next to them. This check already caught one real arithmetic slip before it went further. See .github/workflows/ci.yml.

License

MIT.