Audit Claude session transcripts and generate verdicts from Stop hook signals.
npx skills add https://github.com/Cheggin/request-for-startups --skill review-cheggin
Please help me install this Agent Skill. Skill: review Source: https://github.com/Cheggin/request-for-startups/tree/main/skills/review Command: npx skills add https://github.com/Cheggin/request-for-startups --skill review-cheggin
This skill provides third-party reviews of finished Claude sessions by reading completion signals, detecting patterns such as repeated skill invocations or plan-only runs, and surfacing actionable verdicts to guide improvements.
Run the reviewer against all pending signals by executing the review.mjs script.
High-intent search queries and answers about installing and using this skill.
Auditing Claude session transcripts involves reading completion signals stored during the Stop hook to generate deterministic verdicts for each session. The review process operates out-of-process to avoid bias and surfaces actionable learnings for future prompt refinement.
Reviewing pending session signals requires executing the review script to process payloads stored under .harness/signals. The workflow targets either all pending signals or a specific signal file, analyzing them for patterns like skill-heavy loops or context thrash.
Detecting failure modes in autonomous Claude sessions is done by analyzing completion signals for patterns like repeated skill invocations, plan-only runs, and duplicates. The process reads transcript metadata to identify these loops and writes formal verdicts to a metrics directory.
The Claude session review process enforces deterministic output paths, emitting metrics to .harness/metrics and appending actionable learnings to .harness/learnings/knowledge.md. This structured storage feeds back into chain configuration to guide improvements.
Analyzing a specific signal file instead of all pending Claude session signals is supported by the review workflow. You can target a single signal payload stored under .harness/signals to process individual sessions and generate a verdict without reviewing all pending files.
Using an out-of-process workflow to review Claude sessions ensures a simple, deterministic review that avoids bias. Operating externally from the active sessions allows the reviewer to objectively read completion signals and generate accurate verdicts without session interference.
its-BasudebaExtract reusable development patterns and debugging insights from Claude Code session transcripts.
vuxsoul-uiCoordinate sub-agents to spawn, monitor, and manage multi-step prompt workflows.
ROYCE-8425Extract reusable patterns from Claude Code sessions and save them as learned skills.
acefrost511Analyze user feedback and session insights to update skill definitions.
