agent-execution-discipline

What problem does it solve?

It prevents AI and agent-assisted code changes from being accepted without concrete evidence, verified diagnosis, and repeatable execution discipline.

Core Features & Use Cases

• Evidence-based completion: requires command outputs, validator results, or artifact diffs before claiming completion.
• Verified-cause diagnosis: enforces symptom → hypothesis → method → specific verified cause (no “probably the environment”).
• Failure route repair: after two failures of the same approach, mandates a route change (no third same-path retry).
• Same-pattern scan before local fixes: scans the codebase for the same defect pattern before applying a local fix.
• Reuse and placement rationale: requires rationale for any new structure, preventing utility sprawl and boundary drift.
• Proactive closure package: every handoff/closure includes boundary, validation results, residual risk, and the next target.

Quick Start

Use the agent-execution-discipline skill when an AI is proposing or finalizing a non-trivial code change and you want it to produce an Execution Discipline Report with evidence, verified cause (if diagnosing), and a closure package.