repo-wiki
Repo Wiki
Overview
Create and maintain a structured knowledge base from the codebase. Primary goal: help LLMs and coding agents work faster, with less searching, less repeated explanation, and better architectural grounding. Output should work as lightweight context engine: concise enough to scan, specific enough to act on, and kept in sync as repository changes.
This skill is for LLM-facing internal engineering knowledge such as architecture notes, module pages, PRD-style specs, integration maps, and change-aware wiki upkeep. Human readability matters, but agent efficiency comes first. It is not for public marketing docs or end-user documentation.
When to Use
- Generate an initial wiki for a repository, package, service, or app
- Refresh docs after a PR, commit, refactor, migration, or incident fix
- Create or maintain architecture notes, module pages, feature specs, and PRD-style docs
- Delete stale knowledge when code, routes, jobs, packages, or configs disappear
- Prepare better navigation context, retrieval context, and decision context for coding agents and LLMs
Workflow
- Choose the documentation root.
- Use
.repo-wiki/as the canonical root for generated knowledge. - Do not repurpose
docs/,spec/, orwiki/as the primary output location for this skill. - If those folders contain relevant human-written documentation, reference them from
.repo-wiki/instead of merging into them.
- Use
- Determine scope.
- For PR or commit workflows, inspect the diff and update changed areas first.
- For a first run, do a broader repository survey.
- Read the current knowledge base before editing it.
- Reuse existing filenames, section names, and conventions when they are still coherent.
- Treat code, config, and tests as source of truth.
- Build or refresh the minimum useful knowledge set.
- Repo purpose and major entry points
- System architecture and boundaries
- Module, package, or service pages
- Important flows, integrations, data paths, and operational constraints
- Feature specs or PRD-style pages for active product areas
- Update surgically.
- Create pages for new systems or concepts.
- Revise pages affected by changed code.
- Delete or rename pages when the underlying code is removed or renamed.
- Add navigation and evidence.
- Link related pages.
- Reference concrete files, directories, APIs, or commands in prose.
- Mark unknowns explicitly instead of guessing.
- Validate the result with the recommended layout and update policy.
Default Output Set
.repo-wiki/architecture/overview.mdfor boundaries and major dependencies.repo-wiki/architecture/data-flow.mdfor request paths, events, pipelines, or state transitions.repo-wiki/modules/<name>.mdfor major packages, services, or subsystems.repo-wiki/features/<name>.mdor.repo-wiki/prd/<name>.mdfor feature intent and implementation status.repo-wiki/decisions/<name>.mdfor notable design decisions when those can be inferred responsibly.repo-wiki/glossary.mdfor domain terms and internal vocabulary when needed
Content Rules
- Prefer explicit observations from code, config, tests, and diffs over speculation.
- Keep pages skimmable: lead with purpose, responsibilities, entry points, dependencies, and risks.
- Write generated docs using the writing style guide.
- Use diagrams only when they materially improve navigation; simple Mermaid is enough.
- Separate current behavior from planned behavior.
- Avoid copying large code blocks into docs.
- Highlight mismatches between docs and code instead of silently rewriting product intent.
PR and Commit Update Loop
- Inspect changed files and map them to existing wiki pages.
- Update overview pages if system boundaries, dependencies, or major flows changed.
- Update or create module and feature pages for the touched areas.
- Delete or rename stale pages for removed or renamed code.
- Refresh the root index so new pages are discoverable.
- Call out unresolved questions at the end of the affected page when needed.
Completion Checks
- Every changed subsystem in the PR or commit is reflected in at least one page.
- Removed code no longer has live documentation pretending it still exists.
- Specs distinguish shipped behavior, intended behavior, and unknowns.
- The result makes the codebase easier for an agent to traverse.
Boundaries
- Do not document every file individually unless the repository is tiny.
- Do not restate source code line by line.
- Do not leave empty placeholders without context.
- Do not invent undocumented behavior to make the wiki feel complete.
References
More from bohdan-shulha/skills
qt-qml-attachments
>-
17commit-jira
Create and execute Git commits with a required JIRA-style subject and required body. Use when the user asks to commit changes, write commit messages, or prepare a clean commit and expects a ticket-prefixed subject. Always run a pre-commit safety guard first, abort on log files or untracked high-risk binary extensions, then stage with `git add . -A`.
10claude-cli-agent-protocol
>-
10commit
Create and execute Git commits in Conventional Commits format with a required body. Use when the user asks to commit changes, write commit messages, or prepare a clean commit. Always run a pre-commit safety guard first, abort on log files or untracked high-risk binary extensions, then stage with `git add . -A`.
9plan-b
Create a concise plan for coding/build tasks, especially when users say "I want to make..." or "I need to do...". Use to generate a final plan by aggregating multiple subagent plan variants, include ASCII diagrams/flows/UI mockups when possible, and follow the create-plan style (read-only scan, minimal questions).
8