Hackathon project for an autonomous coding harness that accepts a repository and a natural-language objective, understands the codebase, plans work, edits code, verifies the result, recovers from failures, and returns evidence.
- Product Requirements Document
- 4-person execution plan
- Parent and sub-issue breakdown
- Contribution rules
- Implementation backlog
Repository + Objective
↓
Intent / Goal Analysis
↓
Repository Intelligence
↓
Context Assembly
↓
Planning
↓
Execution + Tools
↓
Verification
┌────┴────┐
PASS FAIL
│ │
Audit Recovery
│ │
└────┬─────┘
↓
Evidence-backed result
The MVP exposes one user workflow rather than separate build, fix, audit, or refactor modes. The harness infers the strategy from the objective.
make runs the full pipeline: venv setup, repository hygiene, lint, the test suite, the judge demo scenarios, and a live TUI session on the demo fixture.
makeEach stage is also available as a standalone target:
make setup— create.venv(viauv) and install dev dependencies; idempotentmake hygiene— repository hygiene checks (./scripts/ci.sh)make lint—ruff check .make test— the pytest suitemake demo— demo scenarios A/B/C with evidence reports underruns/demo/make tui— live TUI session ondemos/fixtures/feature_repo; the M1 pipeline stops after repository inspection, soPARTIAL(exit code 1) is the expected outcome theremake clean— remove run artifacts and cachesmake help— list all targets
Pull requests and pushes to main run .github/workflows/ci.yml. The workflow validates repository hygiene and automatically runs Python and Node checks when those project files appear.
Tags matching v* run .github/workflows/release.yml, re-run the repository checks, create a source archive and SHA-256 checksum, and upload them as release artifacts. A deployment target is intentionally not assumed until the runtime destination is chosen.
Run the same checks locally with:
./scripts/ci.sh