Reference

Full Reference Index

This page is the entry point into the full reference material, split across several pages so nothing requires scrolling through one giant file:

• Skills Reference — every skill, grouped by Setup & Planning, Execution, Status & Review, and Committing & Maintenance.
• Agents — Scrum Master, Developer, Tester: roles, ownership, and responsibilities.
• Hooks — session lifecycle hooks and what they run.
• MCP Tools — Model Context Protocol tools available in a Claude Code session.

The rest of this page covers the repo’s directory structure and the inter-agent communication contract (the typed “sender object” every agent call carries).

---

Per CLAUDE.md’s Source of Truth rule, skills/, agents/, hooks/, scripts/, and templates/ at the repo root are canonical — this is where every feature is created and edited. .agents/ and .claude/ are generated build outputs (populated by j:self-sync in this monorepo, or by the npm package’s postinstall hook in a consumer project) and are never hand-edited.

skills/                    ← Canonical skill source — invoke with j:<name> (see CLAUDE.md)
├── brainstorm/
├── btw/
├── clearify/               (alias: wtf)
├── close-story/
├── commit/
├── continue/
├── convert/
├── deep-dive/
├── dev-done/
├── distribute/
├── do/
├── doc/
├── doc-sync/
├── dooo/
├── error/
├── evaluate/
├── examplify/
├── help/
├── idea/
├── improve/
├── init/
├── j-init/                 (Copilot-collision-safe alias of init)
├── jbp/
├── jenga/
├── jenga-permission-level/
├── lgtm/
├── mirror-public/
├── pi-plan/
├── proceed/
├── publish/
├── reconcile/
├── reconcile-origin/
├── redo/
├── self-sync/
├── skillify/
├── spinoff/
├── status/
├── strategy/
├── todo/
├── train/
├── uncharted/
└── wtf/                    (alias: clearify)

agents/
├── ai_engineer.md
├── developer.md
├── scrum-master.md
├── scrutiny-agent.md
├── solution-assessor.md
└── tester.md

hooks/
├── on_session_end.sh        ← SessionEnd hook (see above)
├── copilot_session_end.sh   ← Copilot-equivalent session-end handling
├── prompt_router.sh
├── prompt_router_helper.js
├── session_end_helper.js
└── session_end_watcher.sh

scripts/                   ← Shell/JS tooling backing skills, CI, and board operations
├── validate-board.sh        ← Board schema/frontmatter validation
├── with-lock.sh             ← Advisory file-lock wrapper for concurrent board writes
├── write-context-digest.sh  ← Writes resolved_context digests (see Agent Communication Contract)
├── postinstall.js           ← npm install hook — mirrors skills/ + agents/ into .claude/ + .agents/
└── ...                      ← see the directory for the full list (board ops, CI, permission tooling)

templates/
├── SCRUM_BOARD_SCHEMA.md
├── PROBLEM_RAPPORT_TEMPLATE.md
├── EXECUTION_PLAN_TEMPLATE.md
├── EXECUTION_SUMMARY_TEMPLATE.md
├── USER_INSTRUCTIONS_TEMPLATE.md
├── CHANGELOG_TEMPLATE.md
├── JENGA_CONFIG_TEMPLATE.json
├── SKILL.md               ← Legacy/simple skill scaffold (predates SKILL_TEMPLATE.md)
├── SKILL_TEMPLATE.md
├── agent-context.md.tpl
├── copilot-instructions.md.tpl
└── permission-levels/       ← level-1-locked.json … level-5-unrestricted.json (see j:jenga-permission-level)

.agents/ and .claude/      ← Generated mirrors — never hand-edited.
                              In this monorepo: refreshed by j:self-sync.
                              In a consumer project: `npm install @jenga-ai/agent` copies only
                              skills/ and agents/ here; hooks/, scripts/, and templates/ are
                              sourced live from node_modules/@jenga-ai/agent/ at runtime.

project/                   ← Created by j:init inside your software project
├── board/
│   ├── epics/             ← E##_<slug>.md
│   ├── stories/           ← E##_S##_<slug>.md
│   └── tasks/             ← E##_S##_T##_<slug>.md
├── configs/
│   ├── workflow.json      ← Shared constants (statuses, paths, agents)
│   └── test-config.json   ← Test tool stack (owned by Tester, user-approved)
├── data/
│   └── baselines.json     ← Analytics baselines (owned by Tester)
├── documentation/
│   ├── plans/             ← Pre-execution plans by Developer
│   └── summaries/         ← Post-execution summaries by Developer
├── ideas.md                ← Lightweight idea log (j:idea)
├── instructions/          ← User-action-prerequisite instructions (Developer-written)
├── queue/
│   ├── scrum_triggers.jsonl
│   ├── developer_triggers.jsonl
│   ├── tester_triggers.jsonl
│   ├── context/            ← resolved_context digests (see Agent Communication Contract)
│   ├── handoffs/           ← Session-boundary handoff files
│   └── project_summary_updates.jsonl
├── rapports/
│   ├── problems/          ← Problem rapports (Developer + Tester)
│   └── analysis/          ← Analysis rapports (Tester)
├── logs/
│   └── events.json        ← Append-only inter-agent event log
└── PROJECT_SUMMARY.md     ← Project source of truth (owned by Scrum Master)

CHANGELOG.md                ← Created by j:init at the project's repo root; maintained by j:publish

---

---

Every call between agents must include a sender object. This is the typed contract that carries context across boundaries.

{
  "sender": {
    "agent": "<scrum-master | developer | tester | orchestrator>",
    "session_id": "<session id from Claude Code>",
    "task_id": "<E##_S##_T##>",
    "story_id": "<E##_S##>",
    "epic_id": "<E##>",
    "date": "<ISO 8601 UTC timestamp>",
    "paths": ["<commit SHA 1>", "<commit SHA 2>"],
    "worktree": "<absolute path to task worktree>",
    "resolved_context": "<optional: path to a digest file under project/queue/context/>"
  }
}

Rules:

• Every agent logs every incoming sender object to project/logs/events.json as its first action
• The Tester rejects any call with missing required fields (responds with "error")
• The paths array contains commit SHAs from the Developer’s commits for this task
• The worktree field is the absolute path to the isolated git worktree created by the Developer
• The resolved_context field is optional: a size-capped digest (under ~100 lines/a few hundred tokens) of what the sending agent already resolved — relevant schema fields, skill precedent, prior decisions — written via scripts/write-context-digest.sh to a unique file under project/queue/context/. It’s a starting point only, never a restriction — the receiving agent may still read full source files when the digest doesn’t cover what it needs.

Required fields by agent:

Field	Scrum Master	Developer	Tester
agent	✅	✅	✅
session_id	✅	✅	✅
task_id	✅	✅	✅
story_id	✅	✅	✅
epic_id	✅	✅	✅
date	✅	✅	✅
paths (commit SHAs)	❌	✅	✅
worktree	❌	✅	✅
resolved_context (optional)	✅	✅	❌