agentsclimarketplace

Maintain docs

Skill pikacss/pikacss/.claude/skills/maintain-docs

Orchestrate PikaCSS documentation maintenance for content drift, missing pages, heading conformity, and API reference generation. Use when: (1) analyzing docs coverage or stale content, (2) code changed and docs may need updates, (3) creating or updating guide/reference/config/plugin pages, (4) updating package READMEs, (5) regenerating API reference pages. This skill is executed directly by the main agent and uses maintain-docs-review for quality review. Translation/zh-TW work is out of scope for this version.From its SKILL.md

Install
npx -y skills add pikacss/pikacss --skill maintain-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

  • 7 stars7 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.

SKILL.md

9.1 KB, as published. Nobody here has run it

maintain-docs

Orchestrate PikaCSS documentation maintenance. English pages only — translation is archived for future work.

Source of Truth

Two reference files define the docs contract:

FilePurpose
references/content-architecture.mdPage inventory — every page, its section, headings (H2/H3/H4)
references/writing-guidelines.mdHow to write — frontmatter, prose style, links, examples, quality checklist

Sidebar and nav are defined in a single VitePress config module:

FilePurpose
docs/.vitepress/sidebarAndNav.tsExports sidebar and nav constants consumed by config.ts

Keep sidebarAndNav.ts in sync with content-architecture.md when pages are added or removed.

Read both references before creating or updating any documentation page. Before drafting content, read the exact source files that will back the page's relatedSources. Use those files to verify public config shapes, concrete CSS variable names, literal layer weights or order values, and any integration or build-tool behavior claims. When a page teaches runtime or config behavior through reusable, behavior-sensitive examples, prefer checked docs/.examples imports over ad-hoc fenced code blocks so docs-scoped validation can verify the snippet through Vitest-backed example tests and/or docs typecheck. Inline fenced code is still acceptable for source-backed signatures, compact config skeletons, module augmentations, or other explanatory snippets where a standalone fixture would add little validation value.

Templates

templates/pages/ contains one markdown skeleton per page, organized by section. Each template provides standard headings and short guidance comments.

Template path maps 1:1 to docs path:

  • templates/pages/getting-started/setup.md → docs/getting-started/setup.md

templates/task.schema.json defines the task file schema produced by analyze.

Commands

All commands use monorepo-level pnpm scripts:

CommandPurpose
pnpm maintain-docs:analyzeCompare templates against existing docs pages. Produces one .task.json per page in .maintain-docs/tasks/.
pnpm maintain-docs:gen-apiGenerate API reference pages from exported surfaces and JSDoc. Reports JSDoc coverage gaps to stdout.

Scripts live under scripts/maintain-docs/ and use workspace-level devDependencies.

Workflow

1. Analyze

Run analyze to produce task files:

pnpm maintain-docs:analyze

This checks every template against its corresponding docs page:

  • Page existence — does the docs page exist?
  • Heading conformity — do H2/H3 headings match the template?
  • Frontmatter validation — are required fields present and valid?
  • Next section — does the page end with ## Next?

Output: one .task.json per page in .maintain-docs/tasks/, plus a stdout summary.

analyze is structural only. It does not prove that examples use valid public shapes, that automatic behavior claims are scoped correctly, or that relatedSources are precise enough.

2. Review Task Files

Read task files to identify work:

ls .maintain-docs/tasks/

Each task file contains:

  • status — missing / outdated / ok
  • issues[] — specific problems found
  • templatePath — the matching template for scaffolding
  • docsPath — the target docs file

3. Create or Update Pages

For missing pages:

  1. Read the template from templatePath.
  2. Read references/writing-guidelines.md for authoring rules.
  3. Read the exact source files that will go into relatedSources before drafting examples, config tables, or behavior claims.
  4. Fill in content following the template structure and guidance comments.
  5. Add proper frontmatter with all required fields.
  6. End with a ## Next section.

For outdated pages:

  1. Read the task file issues[] to understand what needs fixing.
  2. Fix heading structure, frontmatter, or ## Next section as needed.
  3. Apply content quality rules from writing-guidelines.md — check for duplicate code-groups, missing :::tip containers for double-layer keys, undocumented API variants, stray ## Intro headings, invalid public config shapes in examples, over-broad automatic behavior claims, wrapper integrations that fail to name the concrete auto-wiring mechanism, ordering claims that omit literal source-backed layer values, drifted placeholder names across related plugin-authoring pages, unvalidated fenced code where a checked docs/.examples fixture should be preferred, imprecise relatedSources, and missing required metadata on non-index pages.
  4. Re-run analyze to confirm the issues are resolved.

For API reference pages (docs/api/*.md except index.md):

docs/api/index.md is a hand-authored overview page and should stay aligned with its template and content-architecture.md like any other docs page.

Package-level API reference pages are always generated by gen-api-docs — never hand-write or manually edit them.

pnpm maintain-docs:gen-api

If JSDoc coverage gaps are reported in stdout, ask the user whether to address them via the maintain-jsdocs workflow before proceeding.

If the generated links are stale (e.g. dead-link build failures), fix the guideLink data in scripts/_skill-shared/index.ts and re-run the generator in scripts/maintain-docs/gen-api-docs.ts. Do not patch the generated markdown directly.

4. Validate

After making changes, re-run analyze to confirm all addressed pages show ok status.

Before handoff, do a brief source-backed self-review: confirm examples use supported public shapes, automatic behavior claims are scoped to the exact integration or option that provides them, wrapper-package docs name the concrete generated file or template when that is what performs the work, ordering claims preserve literal source-backed values such as -1, placeholder plugin names stay aligned across neighboring authoring pages, relatedSources list the exact current source files, and every non-index page has complete required metadata. If a page includes runnable or behavior-sensitive examples, confirm checked docs/.examples imports are used where they add real validation value. Inline fenced code is acceptable for signatures, type declarations, compact config fragments, or other explanatory snippets that are already source-backed and do not benefit from a dedicated fixture.

For example changes, run:

pnpm --filter @pikacss/docs test
pnpm --filter @pikacss/docs typecheck

Check the pages the diff did not touch. When a change is driven by a source commit that altered behavior, analyze and check-contracts cannot tell you that an untouched page's claim about that same mechanism just went stale — both are structural checks. Read the source commit message for behavior nouns (triggers, determinism, ordering, hooks), then grep every page describing that mechanism and verify it against the current wiring, not just the lines the diff removed. This is how ssr-and-production.md kept a false "generated files are rewritten only when the resolved styles actually changed" claim through the #113 review: the page was never edited, but the refactor that prompted the review had rewired what triggers codegen.

Runtime State

  • .maintain-docs/ — gitignored directory for task files and runtime state.
  • .maintain-docs/tasks/ — flat directory of task files, named with -- separators (e.g., getting-started--setup.task.json).

Agent Pairing

  • Execution: main agent uses this skill directly.
  • Review: hand off completed work to maintain-docs-review for quality gate.

Scope Boundaries

  • In scope: English docs pages, examples, package READMEs, API reference generation, nav/sidebar alignment, reading experience quality (content presentation, layout consistency, custom container usage).
  • Out of scope: zh-TW translation — use the maintain-i18n skill (.agents/skills/maintain-i18n/), which owns docs/zh-tw/** and Taiwan-terminology enforcement. Also out of scope: direct editing of generated dist/ or coverage/ outputs.

Sections

Pages are organized into these sections (from content-architecture.md):

SectionTemplate directory
Getting Startedtemplates/pages/getting-started/
Integrationstemplates/pages/integrations/
Customizationstemplates/pages/customizations/
Official Pluginstemplates/pages/official-plugins/
Plugin Developmenttemplates/pages/plugin-development/
Troubleshootingtemplates/pages/troubleshooting/
API Overviewtemplates/pages/api/
API Package PagesAuto-generated by gen-api-docs — no templates

Keep looking

Skills are one crate of 627,801. 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.