skills.rest
Loringtonian
LoringtonianCOMMUNITY
Explore all skills by this author →
𝕏@loringtonGitHub

harness-review

Reviews recent Claude Code session traces to identify friction and propose improvements.

12|2|Updated Jun 8, 2026
Search Tags:#workflow#claude code#agent behavior#harness#self-review#session traces
One-click install
npx skills add https://github.com/Loringtonian/second-brain-template --skill harness-review-loringtonian
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harness-review
Source: https://github.com/Loringtonian/second-brain-template/tree/main/.claude/skills/harness-review
Command: npx skills add https://github.com/Loringtonian/second-brain-template --skill harness-review-loringtonian

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you find and fix friction in the AI harness itself by reviewing recent Claude Code session traces for repeated mistakes, ignored rules, misfired skills, and unnecessary permission prompts.

Core Features & Use Cases

  • Trace Review: Scans recent session logs to surface concrete moments where the agent struggled or deviated from intent.
  • Pattern Synthesis: Groups repeated friction into clear findings with evidence and the specific rule or intent they conflict with.
  • Gated Improvement Loop: Drafts a proposal first, then waits for explicit owner approval before any change is applied.
  • Use Case: A weekly self-review session can reveal that the agent keeps choosing the wrong folder, missing a skill trigger, or asking for the same permission too often.

Quick Start

Use the harness-review skill to inspect the last seven days of Claude Code traces and draft a finding-by-finding proposal for owner review.

Frequently Asked Questions about harness-review

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

FAQPage Schema
How do I review Claude Code session traces to fix agent behavior?▼

Review Claude Code session traces by scanning recent logs to identify moments where the agent struggled, ignored rules, or misfired skills. The harness-review process groups these frictions into evidence-based findings for owner review.

What causes unnecessary permission prompts in Claude Code workflows?▼

Unnecessary permission churn in Claude Code workflows stems from harness-level friction. Analyzing session traces identifies repeated permission prompts, allowing you to propose targeted improvements that reduce interruptions and streamline agent behavior.

How do I stop my AI agent from ignoring repository guidance rules?▼

Stop your AI agent from ignoring repository guidance by comparing session intent against defined rules. Synthesizing deviations from trace logs enables you to draft gated proposals that correct the harness behavior before applying edits.

Can I automate weekly self-review of my AI agent's performance?▼

Automate weekly self-review by analyzing the last seven days of Claude Code traces, identifying patterns like repeated mistakes or weak sub-agent prompts, and drafting a finding-by-finding proposal for explicit owner approval before any edits.

Does harness-review apply changes to my workspace automatically?▼

No, harness-review does not apply changes automatically. It uses a gated improvement loop that drafts a proposal first, requiring explicit owner approval before any targeted improvements or edits are applied to your workspace.

When do I need to use trace chunking for session reviews?▼

Use trace chunking when reviewing large volumes of Claude Code session logs. It breaks down traces into manageable segments for parallel observation, ensuring accurate pattern synthesis of agent behavior and friction points.

Related Skills

Explore More Software Engineering Skills→
hendrax5hendrax5

audit

Returned successfully.

Community
Intermediate
twikustwikus
3

sync-with-claude

Synchronize local ~/.claude configuration folders to the repository with rsync --delete.

Community
Intermediate
coco-researchcoco-research
264

brain:update

Scan conversations and write new decisions, tasks, entities, and events to a project brain database.

Community
Advanced
sky770825sky770825

session-cleaner

Convert raw OpenClaw and Clawdbot session JSONL logs into markdown transcripts.

Community
Intermediate