/rockets-from-doc

Rockets From Doc Command

Single command to go from spec to working project.

Usage

/rockets-from-doc examples/PRD.md
/rockets-from-doc /path/to/spec.md
/rockets-from-doc docs/requirements.md into ./apps/api

Execution Flow

Step 1: Plan (rockets-planner agent)

Invoke the rockets-planner agent with the spec document. The planner:

1. Reads the spec and extracts entities, fields, relations, ACL, non-CRUD modules
2. Produces a human-readable plan (shown to user)
3. Writes .rockets/plan.json (machine-readable, for orchestrate.js)
4. Waits for user approval before proceeding

Do NOT duplicate planning logic here. The planner is the single owner of plan creation.

Step 2: Bootstrap (if greenfield)

If no project exists at the target path:

node skills/rockets-project-bootstrap/scripts/bootstrap.js --dest <target>

Verify: package.json exists, src/ directory present, dependencies installed, database accessible.

Step 3: Generate (Team or Sequential)

If Agent Teams enabled AND plan has 3+ entities: → Form team per CLAUDE.md > Agent Teams section → Lead creates tasks with blockedBy from topological sort → Teammates self-claim and execute → All CRUD generation costs 0 tokens (pure scripts)

Otherwise (sequential fallback):

node skills/rockets-orchestrator/scripts/orchestrate.js \
  --project <project-path> \
  --spec .rockets/plan.json

→ Topological sort + wave-based execution (0 tokens) → Quality gates: stops on validation failure → Resume on failure: orchestrate.js --project <path> --resume

Step 4: Business logic (non-CRUD)

For modules listed in plan.json.nonCrud[]:

CRITICAL: Non-CRUD services MUST NOT use DataSource or repositories directly (non-model services must consume model services, never repositories). Inject CrudService/ModelService from CRUD modules. For aggregation, use getMany() + in-memory processing. Only exception: DataSource.transaction() for transaction boundaries (workflow services use model services for entity access except DataSource.transaction()).

1. Use rockets-business-logic skill
2. Follow patterns from development-guides/BUSINESS_LOGIC_PATTERNS_GUIDE.md
3. State machines → Pattern 1 (status enums, transition maps, history entities)
4. Custom endpoints → Pattern 2 (approve, cancel, assign, complete)
5. Workflows → Pattern 3 (cross-entity orchestrations)
6. Event automation → Pattern 4 (listeners)
7. Notifications, file uploads, external APIs → Patterns 5-7

Step 5: Validate + Report

# Structural + build validation
node skills/rockets-crud-generator/scripts/validate.js --project <path> --build

# Endpoint smoke test (optional but recommended)
node skills/rockets-runtime-diagnostics/scripts/smoke-test-endpoints.js --project <path>

Step 6: Output report

Return:

1. Entities generated (from orchestrator output)
2. Non-CRUD modules implemented
3. Build status (pass/fail)
4. Smoke test results (if run)
5. Files created/changed
6. Open risks or follow-up tasks

Token Economy

Step	Tool	Tokens
Plan	rockets-planner (opus)	~15k
Bootstrap	script	0
CRUD generation	orchestrate.js (scripts)	0
Business logic	agent (sonnet)	~10k/module
Validation	validate.js (script)	0
Smoke test	smoke-test-endpoints.js (script)	0

Pattern Gates

Enforced automatically by validate.js + hooks/hooks.json. See CLAUDE.md Mandatory Engineering Rules for the full list.

References

• agents/rockets-planner.md — owns plan creation
• skills/rockets-orchestrator/SKILL.md — owns CRUD orchestration
• skills/rockets-crud-generator/SKILL.md — owns file generation
• skills/rockets-business-logic/SKILL.md — owns non-CRUD patterns
• development-guides/ROCKETS_AI_INDEX.md — guide routing