Menu
Reviews the work just done and updates CLAUDE.md, agent memories, and subagent MD files to reflect any new conventions, packages, resources, or patterns discovered. Run after completing any significant implementation task.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
An example custom skill.
An example custom skill version for demonstration.
Plans and scaffolds Anthropic API endpoints as Terraform resources/data sources without a task management tool. TRIGGER when: user says "implement GitHub issues", references a range of GitHub issues to implement (e.g. "issues 52 to 57"), asks to parallelize implementation of multiple resources/data sources, or says "implement all [API name] APIs" or "work on a new API" (e.g. /v1/skills, /v1/environments). Generates GitHub issues and branches if they to not already exist — one per resource/data source — ready for sequential or parallel implementation.
Plans and scaffolds Anthropic API endpoints as Terraform resources/data sources for Vibe Kanban. TRIGGER when: user says "implement GitHub issues", references a range of GitHub issues to implement (e.g. "issues 52 to 57"), asks to parallelize implementation of multiple resources/data sources, or says "implement all [API name] APIs". Creates Vibe Kanban issues, workspaces, and fires parallel agents — one per resource/data source.
Updates the GitHub PR description for the current branch based on the work done in the current Claude session. Use after any meaningful change to a PR (feature, fix, refactor, docs). Invoked automatically when Claude stops if an open PR exists for the current branch. Also works inside Vibe Kanban worktrees (vk/* branches) by resolving the PR via the VK issue context.
Creates a branch named from the work done, commits all changes with a detailed conventional-commit message, pushes to GitHub, and opens a PR (or updates the existing PR description if one already exists) whose title and body come directly from the commit message.
| name | update-project-knowledge |
| description | Reviews the work just done and updates CLAUDE.md, agent memories, and subagent MD files to reflect any new conventions, packages, resources, or patterns discovered. Run after completing any significant implementation task. |
| model | sonnet |
Keep CLAUDE.md and agent memories accurate after each implementation task.
git rev-parse --abbrev-ref HEAD
git log origin/main..HEAD --oneline
git diff origin/main..HEAD --name-only
Also read CLAUDE.md to know what is already documented.
Cross-reference changed files against CLAUDE.md. Check each category:
| If this changed | Check this in CLAUDE.md |
|---|---|
New *_resource.go or *_data_source.go | "Implemented resources and data sources" list |
New make target in GNUmakefile | "Commands" section |
New Go package under internal/ | "Architecture" section |
| New provider-wide convention | "Provider coding conventions" section |
For agent memories, check if the session revealed a pattern that agents got wrong or a non-obvious project idiom — record it in the relevant agent's memory directory.
Skip anything already documented. Skip anything derivable by reading the code.
Edit only files where something is genuinely missing or stale. Prefer adding to existing sections.
Do NOT:
List each file updated and what was added (one line per change). If nothing needed updating, say so.
Then suppress the Stop hook reminder for this branch:
branch=$(git rev-parse --abbrev-ref HEAD 2>/dev/null)
touch "/tmp/claude-knowledge-done-$(echo "$branch" | tr '/' '-')"