Second Brain

capability-activation-gap-new-skillsagents-must-be-wired-int

2 min read

Capability activation gap — new skills/agents must be WIRED into protocols, not just documented

When a session adds a new skill, agent, feedback rule, or hook, it’s not enough to create the definition file. The capability is ORPHAN until a protocol/pipeline/hook/global-CLAUDE.md section actively triggers it.

Evidence (14-Apr-2026 deep audit caught 5 orphans):

  • 4 new agents (critic, architect, tracer, scientist) at ~/.claude/agents/ — not in protocols.md Phase 0 specialist table. Orchestrator had no reason to dispatch them. Fix: add dedicated user-scope agent table with role + model + dispatch timing.
  • ai-slop-cleaner skill — SKILL.md description said “Phase 3.5 Polish on AI-authored code” but protocols.md Phase 3.5 didn’t list it. Fix: protocols.md Phase 3.5 now mandates invocation.
  • commit trailers feedback file — feedback_commit_trailers.md existed and was indexed in MEMORY.md, but session’s 4 commits showed 0% trailer adoption. No trigger at commit-time. Fix: global CLAUDE.md § COMMIT DISCIPLINE added.
  • ast-grep skill — installed but not referenced in Phase 3 Implement / Phase 3.5 Polish. Fix: protocols.md Phase 3.5 now references for structural refactors.

Activation-check rule (add to setup-curator Verification Checklist): After creating any capability, run:

grep -rn <capability-name> /home/claude/.claude/projects/-root-aj-workspace/memory/ /home/claude/.claude/ /root/aj-workspace/.claude/ 2>/dev/null

If the ONLY results are the capability’s own definition file → ORPHAN. Must wire into:

  1. A protocol phase (protocols.md) — for workflow-triggered invocation
  2. A pipeline (plugin-integration.md) — for skill-chain integration
  3. Global CLAUDE.md (behavioral rule) — for session-wide mandate
  4. A Hookify rule — for event-driven autoloading

Before the capability counts as DEPLOYED.

Pattern: Deploy = Define + Wire + Verify. Any step missing = zero impact.

Graph View

Backlinks