Notre avis
Gère la base de connaissances du projet : liste, recherche, ajout d'entrées, audit mémoire et génération de rapports de progression.
Points forts
- Centralise les patterns, conventions et entités du projet dans un fichier structuré.
- Permet une recherche rapide par terme dans la base de connaissances.
- Offre un audit mémoire automatique pour détecter les doublons.
- Génère des rapports de progression quotidiens ou hebdomadaires à partir des données git.
Limites
- Ne fonctionne qu'avec l'outil Claude Code et sa structure .claude.
- L'ajout et la modification sont interactifs, pas automatisés.
- La base de connaissances n'est pas automatiquement synchronisée avec le code.
Lorsque vous devez documenter et retrouver facilement des motifs récurrents, conventions ou entités importantes d'un projet.
Pour une documentation de projet exhaustive ou versionnée, préférez un wiki dédié ou des fichiers markdown dans le dépôt.
Analyse de sécurité
SûrThe skill only reads/writes local project files (knowledge.json, notes.md) and runs git log for reports. No network, destructive commands, or credential handling. The node script call is safe and within the project’s own scripts.
Aucun point d'attention détecté
Exemples
/knowledge list/knowledge search database connection/knowledge addname: mustard-knowledge description: Manage the project knowledge base — list entries, search by term, add new patterns/conventions/entities, run memory audit, or generate progress reports. Use when asked about /knowledge, project knowledge, or pipeline-captured patterns.
<!-- mustard:generated -->/knowledge - Knowledge Management
Notes, memory audit, reports, and project knowledge base.
Trigger
/knowledge <action> [args]
Actions
| Action | Description |
|--------|-------------|
| list | List all knowledge entries grouped by type |
| search <term> | Search entries by name, description, or tags |
| add | Interactively add a knowledge entry |
| notes [target] | Manage project observations |
| audit | Audit memory for duplicates |
| report <period> | Generate progress report (daily/weekly) |
list
Reads .claude/knowledge.json and displays all entries grouped by type.
Flow
- Read
.claude/knowledge.json(if missing: "No knowledge base found. Run a pipeline or use/knowledge addto start one.") - Group entries by
type(pattern / convention / entity) - Display formatted:
=== KNOWLEDGE BASE ({total} entries) === PATTERNS ({n}) - {name}: {description} [source: {source}] CONVENTIONS ({n}) - {name}: {description} [source: {source}] ENTITIES ({n}) - {name}: {description} [source: {source}] - Show last updated timestamp from newest
updatedAt
search
Filters knowledge entries matching the search term across name, description, and tags.
Usage
/knowledge search <term>
Flow
- Read
.claude/knowledge.json - Lowercase-match
termagainstname,description, and each tag - Display matching entries grouped by type (same format as
list) - If no matches: "No entries matching '{term}' found."
add
Interactively adds a knowledge entry by prompting the user, then calls knowledge-update.js.
Flow
- Prompt: "Type? (pattern / convention / entity)"
- Prompt: "Name?"
- Prompt: "Description?"
- Prompt: "Tags? (comma-separated, optional)"
- Build JSON payload and pipe to script:
echo '{"type":"...","name":"...","description":"...","source":"manual","tags":[...]}' \ | node .claude/scripts/knowledge-update.js - Confirm: "Knowledge entry '{name}' saved."
notes
Manages persistent project observations injected into agent context during pipelines. These files are NOT overwritten by /scan.
Targets
| Target | File | Scope |
|--------|------|-------|
| (no argument) | — | Lists all notes files |
| {subproject} | {subproject}/.claude/commands/notes.md | Subproject agent context |
Monorepo: discover targets from pipeline-config.md Agents table or Glob */.claude/commands/notes.md.
Single repo: target is root → .claude/commands/notes.md.
Flow — List (/knowledge notes)
- Read each notes file
- Show summary: which exists, number of observations
- Ask: "Which notes do you want to edit?"
Flow — Edit (/knowledge notes <target>)
- Resolve target to file
- Show current content
- Ask: "What do you want to add, change, or remove?"
- Apply edits
Rules
- NEVER add
<!-- mustard:generated -->to these files - Keep language consistent
- Observations = concise and actionable
audit
Compares auto-memory against CLAUDE.md and skills to detect duplicated information. Outputs report — NEVER auto-edits.
Flow
- Read user's MEMORY.md
- Read all
{subproject}/CLAUDE.md+{subproject}/.claude/commands/guards.md - Present report:
=== MEMORY AUDIT === Total sections: {n} Unique (KEEP): {n} Duplicate: {n} DUPLICATES FOUND: - "{section}" <-> {source_file}: {similarity}% RECOMMENDED PRUNED VERSION: {pruned content} === - NEVER auto-edit — user decides
report
Generates progress reports from git data.
Usage
/knowledge report daily or /knowledge report weekly
Daily Report
git log --oneline --since="00:00" --until="23:59"
git diff --stat HEAD~10
Output: Summary, commits by type (feat/fix/chore), modified files by project, highlights, pending items.
Weekly Report
git log --oneline --since="1 week ago"
git diff --stat @{1.week.ago}
git shortlog -sn --since="1 week ago"
Output: Executive summary, metrics table, implemented features, bugs fixed, changes by project, next week planning.
Rules
- Use real git data only — do not invent commits
- Categorize commits by type and project
Rules
- knowledge.json is persistent — never deleted by session-cleanup
addand pipeline capture both call the sameknowledge-update.jsscriptsearchis case-insensitive- Always show entry count in list/search output
ULTRATHINK
Generateur de Documentation API
Documentation
Genere automatiquement de la documentation API OpenAPI/Swagger.
Rédacteur Technique
Documentation
Rédige de la documentation technique claire selon les meilleurs style guides.
Créer des entrées de journal avec timestamps
Documentation
Crée des entrées de journal horodatées pour documenter les découvertes, réunions, revues de code, articles ou événements significatifs. Structure les informations avec un résumé, un type d'événement, des sources et des tags, puis enregistre le fichier dans `docs/log/` avec un nom basé sur la date et un slug descriptif.