agentsclimarketplace

Docs

Skill justinramos101/agent-skill-kit/skills/.experimental/docs

Battle-tested Agent Skills for coding agents — source-grounded, failure-driven heuristics that audit and design real surfaces. Install with npx skills.

Install
npx -y skills add justinramos101/agent-skill-kit --skill docs

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use for documentation as a product surface, in either direction. Existing docs: audit or debug a README, CHANGELOG, quickstart, onboarding, API reference, examples, contributor guide, help center, or an agent-readable file like llms.txt that also serves human readers — friction, drift, staleness, findability, audience conflict. New docs: design the IA, page structure, source of truth, examples strategy, and versioning for a docs surface before it exists, or define how docs quality is measured — telemetry, CI gates, freshness checks, retrieval evals. Triggers: 'audit our docs/README/help', 'our quickstart fails', 'review our changelog', 'design our docs IA', 'how should we structure the quickstart/reference', 'set up docs CI gates or freshness checks', 'define docs metrics'. Do NOT use for API/SDK/CLI friction beyond the docs (install, error messages, flags), for prose craft on a piece of writing that is not a docs surface, or for standalone agent-native docs and retrieval design.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.4 KB, as published. Nobody here has run it

Docs

Documentation as a product surface, both directions — infer which from the request; never ask:

  • Existing docs → audit or debug. Load references/auditing.md.
  • New or reshaped docs → design the surface or define its measurement. Load references/designing.md.
  • Surface identifiable — README, quickstart, changelog, examples, API/tool reference, contributor docs, help center, dual-audience agent file? Also load references/surfaces.md.

Core principle: match the page to the audience's job and keep it true — and decide the source of truth before writing. Every finding or design choice names a concrete audience (developer, end user, or agent reading a human doc) and its cost; every fact class gets one canonical artifact — other renderings derive from it or are gate-checked.

Invariants

  • Never average conflicting audiences. When a fix helps one audience and harms another, name both and resolve via one source/many renderings, visible-plus-machine-readable, a scope split, or a measured exception — never a taste compromise. No hidden criticals: irreversible consequences, auth scopes, and retry safety never live only in hover or screenshots.
  • Mode purity. One page, one job — tutorial, how-to, reference, or explanation. Mixed-mode pages are findings (audit) or rejected shapes (design).
  • Measurement is a contract. Every metric names its signal, threshold, owner, and action; a dashboard that cannot trigger work is not measurement.

Workflow

  1. Assess. Existing or new? Which intent — audit / debug / design / measure — which surface and audience? State the inference; proceed.
  2. Load only the matching reference(s).
  3. Apply the named heuristics that fit: gap-source trace, copy-paste proof, source-of-truth test, freshness gate, TTFHW, six-part description. Honor the invariants.
  4. Emit. Audit/debug → findings ranked by user cost, each naming audience, mechanism, and fix. Design/measure → a concrete structure or metric contract with acceptance criteria checkable by reading or running.

Run to completion; state assumptions inline. Ask only on a genuine blocker: no docs to inspect and no surface to design. Stop after one emission — a second pass only if your own checks flag a missed audience conflict; never a third.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.