War Room Schema

A war-room is an isolated execution environment for a single epic or task. Each room is a directory under .agents/war-rooms/ containing structured files that track state, communication, and artifacts.

Directory Layout

.agents/war-rooms/room-001/
├── config.json          # Room configuration
├── lifecycle.json       # State machine definition
├── brief.md             # Assignment description
├── channel.jsonl        # Communication log
├── status               # Current state (single line)
├── state_changed_at     # ISO timestamp of last transition
├── retries              # Current retry count (integer)
├── task-ref             # Task reference (e.g., EPIC-001)
├── audit.log            # Audit trail
├── progress.json        # Progress snapshot
├── TASKS.md             # Sub-task checklist
├── artifacts/           # Output files
│   └── ...
├── skills/              # Room-local skill overrides
│   └── ...
└── memory/              # Room-local memory entries
    └── ...

config.json Schema

{
  "room_id": "room-001",
  "task_ref": "EPIC-001",
  "plan_id": "bc0a7c93bcbf",
  "depends_on": ["PLAN-REVIEW"],
  "created_at": "2026-04-03T11:04:01Z",
  "working_dir": "/path/to/project",
  "assignment": {
    "title": "Build the grid system",
    "description": "Full epic description...",
    "assigned_role": "game-engineer",
    "candidate_roles": ["game-engineer", "game-qa"],
    "type": "epic"
  },
  "goals": {
    "definition_of_done": [],
    "acceptance_criteria": [],
    "quality_requirements": {
      "test_coverage_min": 80,
      "lint_clean": true,
      "security_scan_pass": true
    }
  },
  "constraints": {
    "max_retries": 3,
    "timeout_seconds": 900,
    "budget_tokens_max": 500000
  },
  "status": {
    "current": "developing",
    "retries": 0,
    "started_at": "2026-04-03T11:04:01Z",
    "last_state_change": "2026-04-03T11:04:01Z"
  },
  "skill_refs": [
    "detect-ui",
    "unity-dev-principles",
    "create-lifecycle"
  ]
}

Top-Level Fields

Field	Type	Description
`room_id`	`string`	Unique room identifier (`room-NNN`)
`task_ref`	`string`	Epic or task reference (`EPIC-NNN`)
`plan_id`	`string`	Hash linking to the source plan
`depends_on`	`string[]`	Task refs this room depends on
`created_at`	`string`	ISO 8601 creation timestamp
`working_dir`	`string`	Absolute path to working directory

Assignment Block

Field	Type	Description
`title`	`string`	Short assignment title
`description`	`string`	Full assignment description
`assigned_role`	`string`	Primary role for this room
`candidate_roles`	`string[]`	All roles involved in the lifecycle
`type`	`string`	`"epic"`, `"task"`, or `"review"`

Goals Block

Field	Type	Description
`definition_of_done`	`string[]`	Checklist items for completion
`acceptance_criteria`	`string[]`	Testable acceptance criteria
`quality_requirements`	`object`	Minimum quality thresholds

Constraints Block

Field	Type	Description
`max_retries`	`int`	Max retries before `failed-final`
`timeout_seconds`	`int`	Max time for any single agent run
`budget_tokens_max`	`int`	Token budget cap

Status Block

Field	Type	Description
`current`	`string`	Current lifecycle state
`retries`	`int`	Number of retries consumed
`started_at`	`string`	When work began
`last_state_change`	`string`	Last state transition timestamp

File-Level State

Some state is stored as plain files for easy shell access:

File	Content	Example
`status`	Current state name	`developing`
`state_changed_at`	ISO timestamp	`2026-04-03T11:04:01Z`
`retries`	Integer count	`1`
`task-ref`	Task reference	`EPIC-001`

brief.md

Contains the assignment description in markdown. This is injected into the agent’s prompt as the task context. It is generated from the plan’s epic description.

progress.json

Written by agents to report progress:

{
  "percent": 75,
  "message": "Implemented 3 of 4 tasks",
  "updated_at": "2026-04-03T11:19:34Z"
}

TASKS.md

A checklist of sub-tasks within the epic:

- [x] TASK-001 — Grid dot rendering
- [x] TASK-002 — Coordinate mapping
- [ ] TASK-003 — Content rect calculation

audit.log

Append-only log of all state transitions, agent invocations, and system events. Used for debugging and compliance.