cpv-batch-caching-audit

/cpv-batch-caching-audit — Read-only cache audit across many plugins

The cpv-cache-optimizer-agent's full workflow (audit → fix → re-validate → optional Phase-4 broader refactor) is expensive — each Phase-4 step requires interactive confirmation. For a fleet-wide "which plugins have cache-invalidation findings?" snapshot, this command runs only Phase 1 (audit) per plugin and returns the per-plugin severity table. Apply fixes later with /cpv-batch-caching-optimize.

Same input grammar as the rest of the batch family — single plugin local/URL, marketplace local/URL, list, @listfile, or comma-separated.

You are the orchestrator

You — the model running THIS turn — drive the batch from the main session. You do NOT audit anything yourself.

Step 0 — Resolve arguments

If no target was given, ask plain-text:

Which marketplace, plugin, or list should I cache-audit? Provide a path, URL, or @listfile.

Step 1 — Build the batch plan

BATCH_SPEC="$1"
MAX_PARALLEL=8

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/cpv_batch_orchestrator.py" plan \
  "$BATCH_SPEC" \
  --agent cpv-cache-optimizer-agent \
  --mode batch_audit \
  --max-parallel "$MAX_PARALLEL"

Capture the orchestrator's stdout (PLAN, STATUS_TABLE, SESSION_DIR, PLUGIN_COUNT, DISPATCH_GROUPS). Queue the initial status table for the claude-menu-system Stop hook (emitted post-turn via systemMessage — zero token cost, NEVER printed inline by the orchestrator):

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/print_menu.py" "$STATUS_TABLE"

NEVER print menu inline; the CMS Stop hook emits via systemMessage at turn end.

Step 2 — Dispatch audit agents in parallel

for plugin_index in group:
    plugin = plan.plugins[plugin_index]
    Agent(
      subagent_type: "cpv-cache-optimizer-agent",
      description: "Batch-cache-audit {plugin.display_name}",
      prompt: |
        <context>
        source: /cpv-batch-caching-audit
        mode: batch_audit
        plugin_index: {plugin.plugin_index}
        plugin_path: {plugin.abs_path}
        source_url: {plugin.source_url or "—"}
        display_name: {plugin.display_name}
        session_dir: {plan.session_dir}
        status_path: {plan.session_dir}/plugin-{plugin.plugin_index}.status.json
        </context>

        Run Phase 1 (Audit) of your standard cache workflow ONLY —
        skip Phase 2 (Fix), Phase 3 (Re-validate), and Phase 4
        (Broader refactor). Write per-plugin status JSON to
        `status_path` with the canonical batch_audit schema (see
        agents/cpv-cache-optimizer-agent.md). Audit-only, so set `after`
        equal to `before`:

          {
            "status_symbol": "✓" | "✗" | "⚠",
            "status_label": "clean" | "findings" | "warning-only",
            "before": {"critical": N, "major": N, "minor": N, "nit": N, "warning": N},
            "after":  {"critical": N, "major": N, "minor": N, "nit": N, "warning": N},
            "report_path": "<abs-path-to-cache-audit-report>",
            "notes": "<short summary, e.g. 0 CA-01, 2 CA-04>"
          }

        Return ONE line exactly:

          [plugin-{plugin.plugin_index}] {label}: <C>/<M>/<m>/<n>/<w> (status: {status_path})

        Do NOT render menus. Do NOT apply fixes. Do NOT ask the user
        whether to proceed — this is read-only audit mode.
      run_in_background: false
    )