Menu
Detect and remove contradictions across agent policies before execution.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Rules and strategies for managing agent context window size, avoiding bloat, and preserving signal-to-noise ratio.
Standard patterns for error handling, retry logic, circuit breakers, and graceful degradation.
Multi-step tool workflows via code orchestration to reduce latency, context pollution, and token overhead.
Bind project-specific prompts to local schema and workflow artifacts while keeping the harness core generic and globally reusable.
Operational session protocol for task-scoped leases, reconciliation, checkpoints, inspection, queue promotion, and handoff across long-running work.
Operational session protocol for task-scoped leases, reconciliation, checkpoints, inspection, queue promotion, and handoff across long-running work.
| name | policy-coherence-audit |
| description | Detect and remove contradictions across agent policies before execution. |
Detect and remove contradictions across agent policies before execution.
AGENTS.MDharness_session(action: "begin") is the ONLY way to start/reconcile a task.When two policies conflict, resolve them in this order:
If two rules still conflict after this ordering, rewrite the policy so only one source of truth remains.
| Conflict | Fix |
|---|---|
| “Be autonomous” vs “ask after every completion” | make the ask happen at issue boundaries only, not mid-issue |
| “Keep versioning simple” vs “bump every internal contract major” | separate package version from schema/contract versions and document both |
| “Use HarnessOS only” vs ad-hoc SQLite/file mutation | route all canonical task state through the official Harness tools |
| “Use one question with five options” vs freeform open-ended prompts | keep option 4 as Freeform and move nuance into the option descriptions |