v0.0.8

PRD-native coding loops
for AI agents.

Hal turns product requirements into bounded stories, runs them through Codex, Claude Code, or Pi, and leaves reviewable commits, reports, and state before you merge.

Install Hal

Or inspect the README on GitHub

Proof you can inspect before you run it

hal
$ hal init
✓ Created .hal/ directory with config and templates
$ hal plan "add user authentication"
✓ Generated PRD with 8 user stories
$ hal run
→ US-001: Implement login form
✓ Committed: feat: US-001 — login form
→ US-002: Add JWT middleware
✓ Committed: feat: US-002 — JWT middleware
→ US-003: Wire up logout
✓ Committed: feat: US-003 — logout flow
All stories complete. 3 commits ready for review.

Give agents a workflow, not a wish.

Hal wraps AI coding agents in a loop that plans scope, bounds context, preserves state, and keeps the human in review.

Where Hal fits

Use Hal when the task is clear enough to plan, risky enough to review, and repetitive enough to deserve a loop.

Greenfield feature Code review Refactor
hal run — greenfield feature
$ hal plan "Add a dark mode toggle with system preference detection"
✓ 5 stories generated
US-001: Detect system color scheme
US-002: Add theme toggle UI component
US-003: Persist preference to localStorage
US-004: Apply theme class to root element
US-005: Add transition animation
$ hal run
→ Picking next story...
→ US-001: Detect system color scheme
✓ Implemented and committed
→ US-002: Add theme toggle UI component
✓ Implemented and committed
All 5 stories complete. 5 commits ready for review.

Plan. Validate. Run. Review.

One explicit pipeline from product intent to reviewable changes.

STEP 01

Plan

Run hal plan and describe the feature. Hal asks clarifying questions, then generates a structured PRD.

STEP 02

Convert

hal convert transforms the markdown PRD into structured JSON. The runtime state is ready.

STEP 03

Validate

hal validate checks every story against quality rules. Broken PRDs never enter the loop.

STEP 04

Run

hal run picks the next story, runs implementation, commits reviewable state, and repeats through the planned work.

Bring the agent you already use

Hal supplies the loop. Codex, Claude Code, or Pi supplies the implementation engine.

Claude Code
OpenAI Codex
Pi

Run a small loop.
Review every change.

Install Hal, start with a bounded requirement, and inspect the resulting commits before merging.

Or inspect the README on GitHub