review-knowledge

Review Knowledge

Goal

Maintain knowledge base health and surface entries for review, helping the user internalize accumulated knowledge through periodic reflection.

When to Use

• User explicitly requests a knowledge review (e.g., "review knowledge", "check knowledge base")
• At the end of a significant work phase or project milestone
• When the user wants to understand the current state of knowledge on a topic
• Periodically (e.g., weekly/monthly) to keep the knowledge base healthy

Execution

Delegate the review work to a Sonnet subagent to minimize main context consumption.

1. Determine the review mode from user input:
   • No argument or "health" → health
   • topic:<keyword> → topic with the specified keyword
   • fix → fix
2. Determine the knowledge base path (default: .claude/knowledge/entries/)
3. Precompute deterministic graph facts from the MAIN agent's Bash (subagents run in a sandbox that blocks code execution, so these must run here):

python3 "{plugin_root}/scripts/kb_graph.py" --root .claude/knowledge/entries stats
python3 "{plugin_root}/scripts/kb_graph.py" --root .claude/knowledge/entries lint --json

Notes: pure stdlib, ~1s, no index needed. lint exiting 1 just means it found findings — capture stdout either way. If python3 is unavailable or the commands fail, set both outputs to (unavailable) and continue; the subagent then derives everything by reading files as before. 4. Spawn a subagent with the following configuration:

Agent(
  subagent_type: "general-purpose",
  model: "sonnet",
  description: "review knowledge base",
  prompt: |
    You are a knowledge base review agent. Read the procedure file and follow it precisely.

    ## Input
    ### mode
    {mode}

    ### topic_keyword (only for topic mode)
    {topic_keyword}

    ### current_date
    {current_date}

    ### graph_stats (deterministic, precomputed by kb_graph.py)
    {graph_stats}

    ### graph_lint (deterministic, precomputed by kb_graph.py)
    {graph_lint}

    ## Instructions
    1. Read the procedure file at: {plugin_root}/skills/review-knowledge/procedure.md
    2. Follow the procedure step by step for the specified mode
    3. The knowledge base is at: {project_root}/.claude/knowledge/
    4. Return the full review report in the output format specified in the procedure
)