agent-harness-diagnose

Diagnoses root causes of agent harness drift between documentation and repository reality.

Updated Jun 28, 2026
One-click install
npx skills add https://github.com/JaviMontano/claude-plugins --skill agent-harness-diagnose-javimontano
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-harness-diagnose
Source: https://github.com/JaviMontano/claude-plugins/tree/main/plugins/claude-native-toolkit/skills/agent-harness-diagnose
Command: npx skills add https://github.com/JaviMontano/claude-plugins --skill agent-harness-diagnose-javimontano

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Agent harnesses drift over time: progress files claim features that lack evidence, entries reference deleted features, and documented commands no longer work. This Skill pinpoints exactly why a harness stopped reflecting reality, producing a causal diagnosis instead of a vague warning. ## Core Features & Use Cases - Deterministic drift report: Runs reconcile_harness.py to classify drift into six content-derived buckets (claims without entries, orphan entries, empty evidence, broken evidence paths, unverifiable commands, residual placeholders) with no reliance on timestamps. - Causal triage: Names the probable cause and responsible file for each non-empty bucket, and distinguishes mechanical drift (route to repair) from repeated process drift (route to evolve). - Edge-case handling: Reports a corrupt feature_list.json as its own parse_error bucket and treats freshly scaffolded placeholders as expected creator-completion work. - Use Case: A team's progress.md lists features as done but the repo shows no supporting evidence; run the diagnosis to get a per-bucket cause and the next action for each. ## Quick Start Ask the assistant to diagnose the drift of the harness in a given directory because progress.md no longer reflects the actual repository.

Frequently Asked Questions about agent-harness-diagnose

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagnose drift in an agent harness?▼

Run reconcile_harness.py with the --json flag against the harness directory. It produces a drift report with six buckets, and the diagnosis names a probable cause and responsible file for each non-empty bucket.

What kinds of harness drift can be detected?▼

Six content-derived buckets: claims without entries, orphan progress entries, empty evidence, broken evidence paths, unverifiable documented commands, and residual placeholders. All are derived from cross-referencing file content, never from timestamps.

What is the difference between harness diagnose, repair, and audit?▼

Diagnose only identifies root causes and produces the causal report; it applies no changes. Repair applies the fixes, and audit scores the harness. Requests to apply fixes are routed to repair instead.

What happens when feature_list.json is corrupt?▼

A corrupt feature_list.json is reported as its own parse_error bucket containing the exact parser message, and triage of the remaining buckets stops until the parse error is resolved.

When should harness drift not be diagnosed with this approach?▼

Because detection is content-derived, it cannot detect pure temporal abandonment where files are simply stale but internally consistent. It also does not infer team intent or apply any fixes itself.