Audit Skill by Derailment: The Native Two-Tier Auditor

Harden and upgrade any skill by analyzing real execution friction traces. Diagnose root causes across Structural, Semantic, Operational, and Cognitive dimensions, and directly fix the skill text where it broke.

Supports three operational modes:

• Mode A (Synthetic Subagent): Fast, single-agent simulated execution with standard friction markers.
• Mode B (Live Herdr Fleet Audit): Inspecting terminal scrollbacks and thought traces across existing multi-repo Herdr panes.
• Mode C (Native Two-Tier Loop): Provisioning a dedicated Herdr worker tab, monitoring completion, and dispatching an independent forensic auditor subagent to cross-reference terminal scrollback against physical disk evidence.

When to Use

Use this skill when:

• Testing whether a SKILL.md holds up when an autonomous agent executes a non-trivial, multi-step task ("test my skill", "audit skill by derailment").
• Observing live agent execution in Herdr tabs and diagnosing why an agent drifted, guessed, or halted prematurely ("audit herdr pane w2N:pP", "why did the agent stop?").
• Orchestrating an end-to-end audit: spawning a worker agent in Herdr, toggling Ctrl+O verbose mode, waiting for completion, and delegating forensic analysis to an auditor subagent.
• Hardening an existing skill before release or integration into production fleets.
• Post-edit verifying that a fix to a skill closed the friction it was meant to resolve.