| name | harness |
| description | Harness engineering guide — maps all sensors and feedforward guides, teaches self-correction protocol when a CI sensor fires. Use when a sensor fires, before making code changes, or when setting up agent context for a new task. Triggers: "harness", "sensor fire", "CI failure", "self-correction".
|
| category | web-ui |
Skill: harness
What Is the Harness
Agent = Model + Harness. The harness is the system of feedforward guides and feedback sensors.
- Feedforward (guides): Context, constraints, conventions that prevent errors.
- Feedback (sensors): Automated checks that fire after code changes.
Sensor Response Protocol
When a computational sensor fires:
- Read the full error message
- Classify: test / a11y / security / performance / design
- Apply the minimal fix
- Re-run the specific sensor
- Only commit when the sensor is green
Sensors
| Sensor | Command | Config |
|---|
| vitest | bun vitest run | vitest.config.js |
| roast | bash scripts/roast-scorer.sh | ROAST.md |
| security | bash scripts/security-audit.sh | scripts/security-audit.sh |
| links | bash scripts/link-check.sh | scripts/link-check.sh |
| doctor | bash scripts/doctor.sh | scripts/doctor.sh |
| quality | bash scripts/quality-gates.sh | All of the above |
Steering Loop
When any sensor fires repeatedly (>2 times):
- Identify root cause
- Update the corresponding feedforward guide
- If no guide exists, create one in
.agents/skills/
Rules
- Never commit with a failing sensor
- Always run the full quality gate before opening a PR
- Roast score must be >= 80 before merge
- Document repeated failures in lessons/
