headless

name	headless
description	Build an app from a pre-written PRD file without interactive discovery
disable-model-invocation	true

Build an app from a ready PRD. No interview. Designed for programmatic use by other AI agents or CI systems.

Usage: /headless <path-to-prd> [testing-level]

Parse $ARGUMENTS for:

• First argument: path to PRD file (required)
• Second argument: testing level — fast, standard, or full (default: standard)

---

EXECUTION

Step 1: Validate PRD

Read the PRD file and validate against pipeline/prd-schema.md:

1. All 12 sections present
2. App Name and Slug defined
3. Core Features have required fields (Purpose, Action, Response, Success)
4. Monetization model is valid
5. ASO fields within character limits

If validation fails → list all errors, stop. Do NOT attempt to fix or guess.

Step 2: Create App Directory

apps/<slug>/
└── spec/
    ├── prd.md              ← copy the input PRD here
    ├── plan.md             ← will be generated
    └── testing-level.txt   ← from argument or "standard"

Save the testing level to apps/<slug>/spec/testing-level.txt.

Step 3: Plan

Read pipeline/plan.md — generate 9-section implementation plan from the PRD.

No user interaction. Plan is derived entirely from the PRD.

Step 4: Build

5 milestones sequentially: M1 (Scaffold) → M2 (Screens) → M3 (Features) → M4 (Monetization, if enabled) → M5 (Polish)

After EACH milestone, run the verify loop from pipeline/qa.md (reads testing level from spec/testing-level.txt).

Step 5: Finalize

Final QA → apps/<slug>/ralph/FINAL_VERDICT.md.

Output build complete message.

---

DIFFERENCES FROM INTERACTIVE BUILD

Aspect	Interactive (/build-app)	Headless
Discovery interview	Yes (5-8 questions)	Skipped
Web research	Yes (WebSearch)	Skipped (PRD must include §9)
PRD approval	User confirms	PRD accepted as-is
Testing level	User chooses	Argument (default: standard)
Build process	Identical	Identical
QA	Identical	Identical

Key rule: headless mode trusts the PRD completely. It does not validate market research quality, pricing sanity, or feature feasibility.

---

VALIDATION ERRORS

If the PRD is invalid, output:

PRD VALIDATION FAILED

Errors:
- [error 1: which section, what's wrong]
- [error 2]

Fix these errors in the PRD and retry.
Schema: pipeline/prd-schema.md

Do NOT proceed with an invalid PRD. Do NOT attempt to fix it.

---

PROGRAMMATIC USAGE

Other AI agents can invoke headless builds by:

1. Generating a PRD conforming to pipeline/prd-schema.md
2. Saving it to a file
3. Starting Claude Code in the etnamute directory
4. Sending: "Headless build: [fast|standard|full]"