repo-wiki

name	repo-wiki
description	Generate a complete, professional multi-page project wiki for the current repository, using a bundled template (derived from a scraped product wiki). Output language and page titles/filenames must match the user's query language. Output goes to ./.open_docs.

Repo Wiki

Generate a multi-page Markdown wiki for the current repository, using a built-in template derived from a scraped product wiki.

You do not need the scraped JSON at runtime: the template structure is bundled in this skill under references/.

Non-negotiables

• Grounding: All factual technical content must come from the current repo (code/config/docs). Do not invent modules, features, commands, or paths.
• Language: Output language must match the user's query language (auto-detect). If the user explicitly requests a language, that overrides auto-detection.
• No language handoff: Do not write pages that defer content to another language (e.g. “see English page for details”). The page content itself must be written fully in the target language; keep technical tokens unchanged (paths/commands/identifiers/env vars).
• Professional completeness: Follow the template page set and section checklists to produce a complete, professional wiki (overview → architecture → user guide → developer guide → reference → advanced topics).
• Style mimic (not copying): Mimic the template’s structure, tone, and section patterns. Do not copy third‑party paragraphs verbatim unless the user confirms they have rights to do so.
• Output location: Always write Markdown pages to ./.open_docs in the current project.

If the user asks to reproduce third‑party content verbatim and you are not sure they have rights to do so, refuse to copy verbatim and produce an original, repo-grounded wiki that matches the style/structure.

Template references (bundled)

• Template page set + per-page section checklist: references/wiki-template.md
• Scraped JSON format (optional, only if you want to refresh/compare): references/wiki-json-format.md

Output

• Output directory: ./.open_docs
• Always create ./.open_docs/index.md linking to all pages in order.
• Create one Markdown file per page.
• Each page should include a navigation line that links back to index.md and (when applicable) the previous/next page.

Workflow

1. Determine target language

• Detect from the user's query (e.g., Chinese query → Chinese output; English query → English output).
• If ambiguous/mixed, ask a single clarification question.

2. Plan the wiki using the bundled template

• Use references/wiki-template.md as the canonical structure (pages + required sections).
• Ensure page titles and filenames follow the user's query language, per references/wiki-template.md.
• If the repo clearly lacks a concept (e.g., no API, no CLI), keep the page but mark the relevant sections as "Not found in repo" / "Not applicable".

3. Collect repo evidence

• Read the minimum set of authoritative files per page and keep a per-page "Relevant source files" list:
  • top-level docs (README*, docs/, CHANGELOG*)
  • manifests (pyproject.toml, package.json, etc.)
  • entry points / CLI (src/, scripts/, bin/)
  • configuration files and examples
  • CI/CD (.github/workflows, pipelines) if present

4. Write .open_docs Markdown files following the template

• Keep page order stable (follow the template id order).
• File naming: prefix with the template id to keep ordering (e.g., 1-概述.md, 2-架构设计.md).
• Ensure internal links in index.md point to the generated filenames.
• Add a navigation line near the top of each page that links to index.md and previous/next pages (when present).
• Include a "Relevant source files" block (prefer the <details><summary>...</summary> ... </details> style used in the template).

Optional helper script (deterministic dump)

Use this when you want to scaffold ./.open_docs (page files + index) from the built-in template (no JSON required), then fill in content based on the current repo:

python <skill_dir>/scripts/scaffold_open_docs.py --query "<user query>"

If you have a scraped JSON page plan (optional), pass --input <path-to-json> to extract the page list (the wiki content is still written from the current repo).

This script scaffolds filenames and index.md; it does not copy third‑party paragraphs.