cpv-batch-caching-optimize

/cpv-batch-caching-optimize — Apply cache-invalidation fixes across many plugins

For users who maintain a marketplace and want to apply the cache-invalidation pattern fixes (CA-01..CA-07) across every plugin in one pass, this command fans out one cpv-cache-optimizer-agent per plugin in batch_fix mode. Each agent runs Phase 1 (Audit) → Phase 2 (Fix) → Phase 3 (Re-validate) on its assigned plugin and writes its result back as the per-plugin status JSON.

Phase 4 (Broader cache-aware refactor) is deliberately skipped in batch mode: Phase 4 changes require interactive per-step approval and that doesn't compose with a parallel dispatch (each agent would block waiting for the orchestrator's permission). Run the interactive cache flow from /cpv-main-menu (menu-tree §3.3.3 "Audit + broader refactoring") on a single plugin if you want Phase 4.

Same input grammar as the rest of the batch family.

You are the orchestrator

You — the model running THIS turn — drive the batch. You do NOT optimize anything yourself.

Step 0 — Resolve arguments

If no target was given, ask plain-text:

Which marketplace, plugin, or list should I cache-optimize? 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_fix \
  --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 optimize agents in parallel

for plugin_index in group:
    plugin = plan.plugins[plugin_index]
    Agent(
      subagent_type: "cpv-cache-optimizer-agent",
      description: "Batch-cache-optimize {plugin.display_name}",
      prompt: |
        <context>
        source: /cpv-batch-caching-optimize
        mode: batch_fix
        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) → Phase 2 (Fix) → Phase 3 (Re-validate)
        on your standard cache workflow. SKIP Phase 4 (Broader
        refactor) — batch mode does not allow per-step user
        prompts. Write per-plugin status JSON to `status_path` with:

          {
            "status_symbol": "✓" | "✗" | "⚠",
            "status_label": "clean" | "fixed" | "partial" | "failed",
            "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-final-cache-report>",
            "notes": "<short summary, e.g. fixed 5/5, 0 remaining>"
          }

        Return ONE line exactly:

          [plugin-{plugin.plugin_index}] {label}: fixed=X remaining=Y (status: {status_path})

        Do NOT render menus. Do NOT prompt the user for Phase 4 —
        this is non-interactive batch mode.
      run_in_background: false
    )