What problem does it solve? Coding agents like Claude Code, Copilot, Codex, and Cursor produce inconsistent or unsafe output when the surrounding development harness — instructions, linters, CI tiers, pre-commit hooks, and runtime guardrails — is incomplete, misconfigured, or silently diverging. This Skill audits what your harness has and lacks, diagnoses why a specific rule is not being enforced, and designs a harness from scratch for a new project. ## Core Features & Use Cases - Harness Audit: Scans the repository for inferential guides (CLAUDE.md, AGENTS.md, copilot-instructions.md, skills, prompts) and computational sensors (linters, pre-commit hooks, CI tiers, fitness functions, runtime guardrails), then reports gaps in a guides-by-sensors table. - Rule Enforcement Diagnosis: Walks a four-step checklist — is the rule documented, does a sensor verify it, is the sensor actually running, is it strong enough — to explain why an agent keeps violating a rule. - Harness Design from Scratch: Selects stack-appropriate tools (Python, Node/TS, Swift/iOS, Kotlin/Android, Go) from an adaptation table, defines CI tiers, and establishes a single source of truth for multi-agent instruction files. - Use Case: A team notices their Copilot agent repeatedly writes network calls outside the designated Networking module. Use this Skill to trace the failure to a missing fitness function and convert the prose rule into a deterministic architectural test. ## Quick Start Audit my repository's development harness and tell me which guides and sensors are missing for my coding agent.
