Data Flow

Quick Reference
Execution Model: RARV (Reason → Act → Reflect → Verify)
State Machine: backlog → in-progress → review → done
Storage: JSON file (cm-tasks.json) + .cm/ directory
Pipeline: Skill Chain Engine

RARV Execution Cycle

Every task in Cody Master follows the RARV cycle — a 4-phase execution loop that prevents AI agents from rushing to incomplete solutions.

Phase Details

Phase	What happens	Key skill
REASON	Read working memory, understand context, pick the right approach	`cm-continuity`, `cm-planning`
ACT	Execute the plan — write code, fix bugs, create docs	`cm-tdd`, `cm-execution`, `cm-debugging`
REFLECT	Self-review — did the output match expectations? Record errors.	`cm-code-review`, `cm-continuity`
VERIFY	Run tests, validate, gather evidence	`cm-quality-gate`, `cm-test-gate`

Task State Machine

Tasks flow through a kanban board with validated transitions:

Valid Transitions

From	To	Trigger
`backlog`	`in-progress`	Developer/agent starts work
`in-progress`	`review`	Work completed, needs review
`in-progress`	`done`	Work completed and verified
`in-progress`	`backlog`	Deprioritized or blocked
`review`	`done`	Review approved
`review`	`in-progress`	Review found issues
`done`	`backlog`	Task needs rework

Validation

Invalid transitions are rejected by the API. For example, you cannot move a task directly from backlog to done — it must pass through in-progress first.

Skill Chain Pipeline

Complex workflows are composed as multi-step skill chains. The Chain Engine manages progress through sequential skills:

Judge Agent Decision Flow

The Judge Agent continuously monitors task health and recommends action:

Data Storage Schema

All data is stored in a single JSON file (cm-tasks.json):

json

{
  "projects": [
    {
      "id": "uuid",
      "name": "My Project",
      "path": "/path/to/project",
      "agents": ["antigravity", "claude"],
      "createdAt": "ISO-timestamp"
    }
  ],
  "tasks": [
    {
      "id": "uuid",
      "projectId": "uuid",
      "title": "Add authentication",
      "column": "in-progress",
      "priority": "high",
      "agent": "antigravity",
      "skill": "cm-tdd",
      "order": 0,
      "createdAt": "ISO-timestamp",
      "updatedAt": "ISO-timestamp",
      "dispatchStatus": "dispatched"
    }
  ],
  "deployments": [...],
  "changelog": [...],
  "activities": [...],
  "chainExecutions": [...]
}

Working Memory Files (`.cm/` directory)

.cm/
├── CONTINUITY.md      # Active session state
├── config.yaml        # RARV cycle settings
└── memory/
    ├── learnings.json  # Captured error patterns
    └── decisions.json  # Architecture decisions

Data Flow ​

RARV Execution Cycle ​

Phase Details ​

Task State Machine ​

Valid Transitions ​

Skill Chain Pipeline ​

Judge Agent Decision Flow ​

Data Storage Schema ​

Working Memory Files (.cm/ directory) ​

Related ​