القائمة
Read-only KB discovery with progressive 4-tier loading and --challenge mode for counter-evidence.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Manages a personal knowledge base with learning tracking. Modes: (1) learn from articles, (2) learn about topics, (3) fix errors. Uses markdown links with relative paths for cross-references. Maintains a rolling changelog, validates links, and keeps a markdown-compatible knowledge base. When a note reaches ~500 lines, suggests /kb-compact.
Sets up the knowledge base for a project or initializes the global (god) KB config. Creates the KB config, directory structure, and appends minimal instructions to the project's CLAUDE.md. Run this once when first using KB skills in a new project. Use --global to set up the global read-only KB.
Monitors KB usage patterns and skill health. Activates when KB access tracking detects frequently-read content that could become a skill, or when user corrections suggest a skill is underperforming. Also triggers on `this isn't working right`, `this skill keeps getting X wrong`, `convert this to a skill`, or `what should become a skill`. Tracks cross-session observations via memory.
Opens the KB viewer in the browser. Starts a local HTTP server that renders all configured knowledge bases with working links, fuzzy search, and knowledge graph visualization.
Compacts KB directories — extracts legacy, unifies terminology, splits oversized notes, fixes indexes, reconciles with skill folders. Default: single directory. Use --deep for recursive bottom-up traversal.
Mints skills from KB topics, packages skills into plugins, and prepares them for Cowork. Use when asked to create a skill from KB content, convert to a plugin, package for distribution, or make something cowork-ready. Also triggers on `convert this to a skill`, `package as plugin`, `mint a skill`, or `make this a plugin`.
| name | kb-find |
| description | Read-only KB discovery with progressive 4-tier loading and --challenge mode for counter-evidence. |
| argument-hint | <query> [--challenge] |
| effort | medium |
Progressive 4-tier KB discovery. Read-only.
index.md or config.json — kb_loader handles these; config auto-detects..md files (Tier 2/3/4).[read-only] badge in topic listing) are searchable but not writable. Treat them identically for discovery.Parse query from $0, strip flags. Two modes:
--challenge: find content that contradicts or complicates the query. Score by contradiction potential. Include adjacent topics. Read 1-2 extra notes in Tier 3/4. Skip KB meta-topics unless query is about them.KB_LOADER="${CLAUDE_SKILL_DIR}/scripts/kb_loader.py"
$KB_LOADER --list-topics — tree by @kb-name with [skill] tags.compact) or @kb-name/path. Challenge: also contradicting topics.$KB_LOADER --topic <path> per topic. All calls in one turn.{topic_name, notes[], children[], has_skill}.--topic returns JSON (names/descriptions) for Tier 2 scoring. Empty list → report and stop.
Skip if Tier 1 name+description clearly shows relevance → Tier 3/4. Only scan ambiguous notes.
Skip to Tier 4 if no TOC or note is <100 lines.
offset=16, limit=15.@kb-name/path) → --topic @kb-name/topic.see: followed by @kb/topic) → --topic @kb/topic. These are semantic hints to global KBs.offset to skip prior lines.Return: ## Topic Map (paths), ## Relevant Concept Notes (path + title + content, label concept vs skill), ## Notes (caveats).
Challenge mode output: ## Topic Map, ## Counter-Evidence (path, strength, type, excerpt), ## Summary (1-3 sentences), ## Notes.