agentleFS
openvigletopenviglet/shio/.claude/skills/shio-docs-sync/SKILL.md

shio-docs-sync

Skill3 starsChanged 8 days agoApache-2.0View on GitHub

Validate that Shio's end-user documentation (shio-site/docs, served at shio.viglet.org/docs) is in sync with the code, and report or file every drift. Runs the docs conformance suites (vocabulary, stack, coverage), a reverse code-to-docs report (CLI verbs, MCP tools and resources, batch ops, verify checks, shio.* properties, manifest features no page names), sidebar membership and internal-link/anchor resolution, then a judgment pass over what no instrument reads (defaults, limits, behaviour described in prose, pages that claim to be complete, features shipped since a page was last touched). Use when asked to check, audit, validate or sync the docs against the code, before a release, after a block completes, or when a docs page is suspected stale. Trigger words: docs sync, documentação, shio-site/docs, stale docs, COVERAGE.md, configuration reference, docs drift.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run from the repo root. Three passes, in this order, because each is cheaper and more certain than the next.

    line 12
  2. Distinctive:

    No instrument reads meaning.

    line 54
  3. Very specific:

    Fixing: follow shio-roadmap-docs § end-user docs (write for the user, fold into the closest page, update the COVERAGE row, pnpm run …

    line 85

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Is shio-site/docs in sync with the code?

The pages live in shio-site/docs/src/content/docs/ (SH1262), with COVERAGE.md beside them (excluded from the collection). Writing a page and recording its COVERAGE verdict is shio-roadmap-docs's job; this skill only measures, then fixes or files what it measured.

Run from the repo root. Three passes, in this order, because each is cheaper and more certain than the next.

Pass 1: the gate (docs → code, mechanical)

pnpm conformance:docs

It runs docs-vocabulary (every verb, route, MCP tool/resource, op, --checks group and shio.* property a page names still exists; every registered tool is on the tool table; every section type is named), docs-stack (stated versions match pom.xml/package.json; no retired dependency presented as part of the stack) and docs-coverage (COVERAGE's summary matches its tables). Read shio-conformance-docs.log; never pipe the gate into grep (agents.md, SH354/ SH788). A red here is a confirmed drift with the fix printed: start with it.

The fourth instrument, docs-examples.conformance.mjs, issues the documented requests and needs a running instance (it is in pnpm conformance, not here). Run it only when the question is "does the documented request still work", not "is the name right".

Pass 2: the reverse report (code → docs, mechanical)

node .claude/skills/shio-docs-sync/docs-sync-report.mjs          # add --json or --strict

The gate is blind in this direction: something the product has and no page mentions passes all three suites. The report reads the same populations the vocabulary suite does (cli/conformance/support/product-vocabulary.mjs), so the two cannot disagree about what exists, and prints:

  • declared, named on no page, per population. shio.* properties accept the camelCase form (Spring relaxed binding). manifest feature keys is advisory: pages describe a feature in prose, rarely by its key, so a miss is a question, not a finding;
  • pages missing from the sidebar in shio-site/docs/astro.config.mjs (they build as unlisted routes nobody reaches);
  • hand-typed /docs/... links whose page, heading anchor or public/ file does not exist.

--strict exits 1 on any non-advisory gap; the default is 0, because some gaps are deliberate.

Pass 3: what only a reader can judge

No instrument reads meaning. Spend the reading where drift is likely, not across every page:

  1. What shipped since the page was last touched. git log -1 --format=%cs -- <page> gives the page's date; git log --since=<date> --oneline -- <the code the page documents> lists the commits it may not know about. A feat: there with no COVERAGE row, or a row pointing at an anchor that does not describe it, is drift.
  2. Pages that claim completeness. "Every property below…", "the twelve tools", "all endpoints": a completeness claim is falsified by a single gap from Pass 2 (configuration-reference.md is the standing example: it says it lists every property).
  3. Defaults and limits in tables. For each | key | default | row, compare with the @Value("${key:default}") / @ConfigurationProperties field or application.properties. A name check passes a wrong default.
  4. Behaviour in prose. Status codes, response shapes, CLI flags, console steps. Check the handler or the command's --help, not memory. A ⚠️ COVERAGE row means a page already describes a shape the product left; re-read those first.

Classify, then act

For every gap, decide one of three, and say which:

VerdictAction
Drift: the page is wrong or silent about something a user would do differentlyFix it if the user asked for fixes; otherwise file a task
Deliberately internal (an internal limit, a test-only property)A 🔒 row in COVERAGE.md if the cluster has none; nothing else
False positive of the reportNote it; if it recurs, fix the report rather than re-judging it each run

Filing: prove the gap with a command first (a grep with zero hits, a --help, a handler line), never from reading alone. Search the backlog before adding (mcp__roadkeep__list / section_find), then mcp__roadkeep__add under the docs block (shio-roadmap-docs names it), one task per page or per coherent cluster, not one per missing key. Name the evidence in the line.

Fixing: follow shio-roadmap-docs § end-user docs (write for the user, fold into the closest page, update the COVERAGE row, pnpm run coverage:mirror). Then re-run Pass 1 and Pass 2; a fix that leaves its gap in the report is not done.

Report back

One table per pass: what was checked (counts, printed by each tool), what failed, the verdict per gap, and what was fixed or filed (task ids). State plainly what was not checked (for example, Pass 3 sampled four pages, or docs-examples was skipped for lack of an instance).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.