domains / families / agent-tooling / harness-engineering

harness-engineering

separate

This skill should be used when designing autonomous agent harnesses: research loops, evaluation scaffolds, locked and editable surfaces, durable logs, novelty gates, pruning, rollback, PR preparation, and human approval boundaries.

When to use

• Building autonomous research or experimentation loops
• Designing agent environments with locked metrics and editable code
• Creating PR-producing or background agents
• Evaluating agent safety for autonomous runs
• Adding novelty, ablation, pruning, rollback, or durable logging
• Preventing agents from gaming benchmarks or losing state

When not to use

• General quality gates, regression suites, or outcome metrics without autonomous control surfaces
• Tool schemas, response formats, and recovery errors for harness tools
• Project-level task-model fit, pipeline shape, and cost planning
• Remote sandbox, warm-pool, and hosted session infrastructure

Skill page

Source