Plugin Layer — Skills, Agents, and Hooks

Layer 3 (Experience) — how Claude Code interacts with the governance engine

21
Skills
Guide Claude Code through governance workflows. Each skill calls composite aggregators for data, then structures AI reasoning.
  • standup — morning briefing from audit trail
  • board — flow intelligence (blockers, risks)
  • plan-wave/sprint/cycle — methodology-specific
  • retro-wave/sprint/cycle — evidence-based retro
  • compliance — 5-category score analysis
  • decompose — strategic → technical spec
  • onboard — zero-friction first touch with auto-clone
  • guided-demo — four-act governance walkthrough
  • sandbox-explore — interactive exploration paths
  • sandbox — lifecycle (create, reset, destroy)
4
Agents
Specialized AI roles invoked by skills. Each agent has focused instructions and specific tool access.
  • PM Agent — governance personified. Uses all MCP tools. Reads memory.
  • code-analyzer — maps capabilities to code. Uses Read/Glob/Grep. Opus model.
  • technical-spec-writer — decomposes into tasks. Judges effort/risk/AI. Opus model.
  • spec-reviewer — independent two-stage review. Format + quality. Sonnet model.
2
Hooks
Automated governance checks triggered by tool events. Block or allow based on governance signals.
  • Post-transition — fires on state changes. Checks: did this unblock downstream? Create new blocker? Reach milestone?
  • Post-wave-assignment — fires on assign_task_to_wave. Checks epic integrity + dependency violations.
Data Flow: Skill to Domain Services
Skill
e.g. /ido4dev:standup
→
Composite Aggregator
get_standup_data (1 call)
→
Domain Services
10-12 parallel queries
→
GitHub + Audit
GraphQL + JSONL
←
Aggregated Data
tasks, analytics, compliance
←
AI Reasoning
structured output
Pattern: Single Composite Call
/ido4dev:standup → get_standup_data → returns: container status, tasks, PR reviews, blocker analyses, audit trail (24h), analytics, agents, compliance → AI structures morning briefing
Pattern: Multi-Step Governance
/ido4dev:plan-wave → get_project_status → list_tasks → search_epics → analyze_dependencies → AI composes wave plan
Pattern: Agent Pipeline
/ido4dev:decompose → parse_strategic_spec → code-analyzer (Read/Glob/Grep) → canvas → technical-spec-writer → ingest_spec
Pattern: Hook Governance
start_task fires → post-transition hook checks: unblocked downstream? new blocker? milestone? → "block" (surface to user) or "allow" (proceed)
Composite aggregators are the performance layer: Skills don't make 12 individual tool calls. One aggregator call runs parallel batches internally, ensures data consistency, and reduces round-trips. The skill gets everything it needs in one response, then focuses purely on AI reasoning.