How dagztagz/dagz-scaffold works
Dagz-Scaffold inspects coding-agent work: a user starts a task persist, Grok produces a durable run packet, and the Python scorer and ship gate decide whether it is acceptable. The packet preserves the plan, changed code, command evidence, critic verdict, and score claim. Major branches include task-specific verification contracts, optional safe command replay, and staged-file checks for secrets or live runs. The workflow wiring is supported by the README and sampled harness code; the skill directories are documented capabilities, but their internal behavior was not sampled.
Main entry points: score.py Packet scorer · run.py Persist runner · ship_gate.py Ship gate · tasks Task definition
Components 16
Persist workflow
- Task definition
tasks - Persist guidance
.grok/skills/persist - Persist runner
harness/run.py - Critic guidance
.grok/skills/critic - Run packet
runs
Run inspection
- Packet scorer
harness/score.py - Task contracts
harness/task_contract.py - Score schema
harness/schema/run.schema.json - Evidence replay
harness/replay.py - Score result
Ship gate
- Ship gate
harness/ship_gate.py - Gate guidance
.grok/skills/ship-gate
Other components
- Task author
- Grok Build CLI
- xAI service
- Git index
Connections 20
- Task author → Persist runner: starts persist
- Task definition → Persist runner: supplies task
- Persist guidance → Grok Build CLI: guides persist
- Persist runner → Grok Build CLI: invokes
- Grok Build CLI → xAI service: uses account
- Grok Build CLI → Run packet: writes artifacts
- Critic guidance → Grok Build CLI: guides critique
- Persist runner → Packet scorer: scores run
- Packet scorer → Run packet: reads packet
- Packet scorer → Task contracts: loads contract
- Task contracts → Task definition: reads task
- Packet scorer → Score schema: validates against
- Packet scorer → Evidence replay: optionally replays
- Packet scorer → Score result: produces grade
- Persist runner → Score result: checks grade
- Task author → Ship gate: requests gate
- Gate guidance → Grok Build CLI: guides gate
- Ship gate → Packet scorer: checks run
- Ship gate → Git index: optionally scans
- Ship gate → Score result: reports pass or block