v3.2.0 — 13 agents, 25 hook events, observability layer

What's new in v3.2.0

Agents

• 13 active agents (restored: security-guard, scope-guard, fe-mentor, skill-suggester)
• All 13 agents now have effort field (high/medium/low)
• builder + tester: permissionMode: acceptEdits
• sec-auditor + security-guard: skills: [security-audit]

Hooks (40 hooks, 25 event types — 100% coverage)

• 5 new hook events: TaskCreated, TaskCompleted, InstructionsLoaded, Elicitation, ElicitationResult
• subagent_verify.py: detect empty/low-quality agent output on SubagentStop
• New scripts: task_audit.py, instructions_audit.py, elicitation_guard.py

Observability Layer (NEW)

• metrics_collector.py: aggregate JSONL logs into session/weekly metrics
• weekly_review.py: automated analysis of recurring failures, memory bloat, dead config
• config_audit_scan.py: AgentShield self-audit of settings, agents, CLAUDE.md, MCP configs

Rules

• Verify-Output Principle added to integrity.md
• Inter-Agent File Contracts added to context-loading.md
• New template: memory/templates/adr-ai-errors.template.md

Docs

• All counts aligned: 13 agents, 40 hooks, 25 events, 17 skills
• Plugin/marketplace bumped to v3.2.0

Full stats

• 395 tests passing
• 38% coverage (new scripts pending tests)
• ruff clean, mypy checked