claude-copilot: Command for Claude Code

Extensions Status

Show the configured Knowledge ladder, reusable skills, and active agent extensions. Do not create or modify files.

1. Inspect the real Knowledge ladder

The only authoritative inputs are paths.knowledge_repo and the ordered CC_KNOWLEDGE_REPOS value from cc env (personal → department → organization → foundation). Do not inspect ~/.claude/knowledge/knowledge-manifest.json or ./knowledge-manifest.json unless that path is actually in the ladder.

eval "$(cc env)"
echo "${CC_KNOWLEDGE_REPOS:-}"
echo "${CC_KNOWLEDGE_REPOS:-}" | tr ',' '\n' | while IFS= read -r repo; do
  [[ -z "$repo" ]] && continue
  if [[ -f "$repo/knowledge-manifest.json" ]]; then
    name="$(python3 -c "import json,sys; print(json.load(open(sys.argv[1])).get('name','?'))" "$repo/knowledge-manifest.json" 2>/dev/null)"
    echo "FOUND: $repo (name: $name)"
  else
    echo "MISSING: $repo"
  fi
done

Store each tier's position, path, manifest name, and found/missing state. The tr/while form is portable across bash and zsh.

2. Resolve skills and extensions

cc skill list
cc extensions resolve --all --json

Use the returned JSON as source truth. For every agent it supplies agent, action, matched, type, file, source_repo, description, requiredSkills, missingSkills, and warning. Resolution is nearest-first: the first tier declaring an agent wins; otherwise the base agent runs unchanged.

Interpret action without inventing state:

• apply: the declared extension is usable; show its type and source tier.
• fallback_use_base: no usable extension was selected; the base agent runs.
• fallback_use_base_with_warning: the base agent runs, and the returned warning must be shown.
• fallback_fail: show the returned warning as a blocking configuration error. Do not claim the agent can run.
• no_extension: no tier declares that agent; this is a healthy base-agent state.

requiredSkills and missingSkills are evidence, not suggestions. When skills are missing, list their exact names beside the affected agent. Do not scan directories or recompute whether a requirement is satisfied; the resolver already applied the same contract used during agent invocation.