doom-doc-assistant — Skill Development Conventions

This file governs development of the doom-doc-assistant skill specifically. It extends the root AGENTS.md with patterns specific to content-generation skills for Doom documentation repositories.

Core Principle

Explicit rules and product documentation standards win.

The target repository's explicit rules are authoritative. When a repository is silent, the product documentation standards bundled in this skill are authoritative. Local page samples are useful for placement, ordering, and style, but legacy samples must not override required product documentation standards such as English product content, underscore-only new paths, weight plus queries, or directory index.mdx coverage.

AI usability is a first-class documentation quality target. When the user asks whether documentation is ready for AI writing, review, retrieval, or assistant answering, the skill must explicitly inspect value sources, prerequisites, support boundaries, and cross-page consistency instead of limiting itself to formatting or Doom syntax.

Current Layout

doom-doc-assistant/
├── SKILL.md
├── AGENTS.md
├── CLAUDE.md
├── rules/
│   ├── ai-usable-docs.md
│   ├── best-practices.md
│   ├── common-pitfalls.md
│   ├── content-elements.md
│   ├── core-conventions.md
│   ├── language-style.md
│   ├── markdown-formatting.md
│   ├── mdx-components.md
│   ├── metadata-rules.md
│   ├── terminology-consistency.md
│   ├── terminology-guide.md
│   └── verification-checklist.md
├── templates/
│   ├── arch-template.mdx
│   ├── concept-template.mdx
│   ├── diagnosis-report.md
│   ├── function-template.mdx
│   ├── howto-template.mdx
│   ├── installation-template.mdx
│   ├── intro-template.mdx
│   ├── quickstart-template.mdx
│   ├── release-notes-template.mdx
│   ├── spec-review-report.md
│   ├── troubleshooting-template.mdx
│   └── upgrade-template.mdx
└── references/
    └── regression-cases.md

Keep this file aligned with the real directory tree. Do not leave historical placeholder names here.

Rules And Templates

rules/

• rules/ files store defaults and reusable checks.
• A rule file must never overrule explicit target repository rules.
• Put stable behaviors here, not repo-specific one-off policies.
• If a rule becomes "discover from the repository first," say that directly in the rule file.

templates/

• Templates are structure references, not truth sources.
• Template file names describe content shapes only. They are not canonical category values.
• Templates for new product documentation must include the stable minimum frontmatter contract: weight and English queries.
• Keep author and category out of template frontmatter unless a future explicit repository rule requires them.

Discovery-First Patterns

Prefer runtime discovery over memorized standards for:

• local weight ordering and page placement
• optional frontmatter fields and category values
• MDX component syntax
• link patterns
• page structure and section naming

The preferred sequence is:

1. Read the target repository AGENTS.md when it exists
2. Apply the built-in product documentation standards from rules/
3. Sample sibling pages in the same directory for local ordering, placement, and non-conflicting style
4. Sample adjacent modules when local evidence is too thin
5. Fall back to templates only for structural scaffolding

Workflow Design Expectations

• The skill must lock the assistant-facing output language before Phase 0 when the preference is not explicit.
• This language choice is separate from the repository's documentation content language.
• Assistant-facing output may be English or Chinese, but drafted or revised repository documentation content must always stay in English.
• The skill must classify the documentation layer before planning or drafting:
  • user-facing product doc
  • engineering fact doc
  • versioned validation report
  • known issue tracker
  • evidence index
• The skill must not silently mix document layers. Public product docs should not inherit raw evidence bookkeeping, and engineering docs should not overstate support by copying user-facing prose.
• New product documentation paths must use lowercase letters, numbers, and underscores only.
• New product documentation frontmatter must include weight and English queries.
• For AI-usable product docs, queries should cover user intent plus high-value platform terms, field names, or aliases when they materially improve retrieval.
• For user-facing product docs, the workflow should explicitly consider prerequisite checklists, value-to-field mappings, controller-managed fields, and supported or unsupported boundaries.
• For engineering docs, the workflow should explicitly distinguish released capability, known gaps, known issues, evidence links, and version baselines.
• If a change introduces or revises a prerequisite, field constraint, lifecycle boundary, or limitation, the plan must either propagate it to affected sibling pages or record explicit debt.
• For Doom/Yarn repositories, the post-drafting handoff must run yarn up @alauda/doom and then yarn install before human acceptance.
• Manual acceptance must be handed to a human reviewer who runs yarn dev locally only after local preview prep completes successfully.
• yarn build and yarn translate are never default verification steps. They are separate tasks that require explicit user direction.
• Phase 0 must output a fixed diagnosis contract, not a free-form summary.
• Phase 1 must output a fixed execution-plan contract, not a new-doc-only checklist.
• Phase 1 and the final documentation summary must include a local preview prep and manual acceptance handoff section.
• Review-only or audit-only requests must stay read-only and stop after a findings report unless the user asks for fixes.
• Explicit yarn build or yarn translate requests must use a separate command branch and must not become default documentation validation.
• Explicit yarn build or yarn translate requests should use a fixed Explicit Command Result contract instead of free-form command reporting.
• Manual acceptance handoff must tell the human to run yarn dev only after local preview prep completes successfully.
• Path A must support modifying authoritative existing docs.
• Path B should only be recommended when restructuring is necessary for this round.
• Path C should capture technical debt without silently expanding scope.

Validation Assets

When changing this skill, use references/regression-cases.md as the minimum regression suite. Add new cases when the skill learns a new branch or fixes a new class of failure.

When To Update Which File

What Not To Do

• Do not force the agent to ask for a repository path when the current workspace already identifies the target repo.
• Metadata and legacy samples must not become fake rules: require author or category only when explicit repository rules require them, and do not let legacy naming, missing queries, or missing index.mdx in existing files dictate new product documentation.
• Command handling must stay narrow: do not automatically run yarn dev, yarn build, or yarn translate, and do not tell the human to run yarn dev after a failed yarn up @alauda/doom or yarn install.
• Read-only review or audit requests must stay read-only; do not run local preview prep commands in those routes.
• Do not treat technical debt as automatic evidence that Path B is required.
• Do not let a user preference for Chinese assistant output turn into Chinese documentation content.
• Do not treat "code path exists" as equivalent to "supported," "documented workflow," or "formally validated."
• Do not let engineering-truth documents leak baseline-specific or issue-specific claims into public docs unless those details are necessary for user success and stated at the right abstraction level.