retrospect

Analyze agent session failures and convert verified behavioral problems into enforcement patches.

4|Updated Feb 23, 2026
One-click install
npx skills add https://github.com/npow/claude-skills --skill retrospect-npow
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: retrospect
Source: https://github.com/npow/claude-skills/tree/main/retrospect
Command: npx skills add https://github.com/npow/claude-skills --skill retrospect-npow

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill identifies failures, corrections, and alignment gaps in an agent session so recurring mistakes can be diagnosed and converted into durable improvements.

Core Features & Use Cases

  • Failure Detection: Scans conversations for explicit corrections, tool failures, omitted verification, premature completion, and other behavioral signals.
  • Root-Cause Analysis: Classifies failures across missing, undiscovered, ignored, ambiguous, or incorrect context and task-framing issues.
  • Enforced Improvements: Prescribes fixes across rule cards, stop classifiers, tool gates, pre-commit hooks, skills, and documentation, prioritizing mechanisms that prevent recurrence.
  • Use Case: After an autopilot or deep-qa session, use this Skill to analyze what went wrong, verify whether existing enforcement caught it, and produce targeted patches with a concise retrospective summary.

Quick Start

Ask the retrospect skill to analyze the current session for failures and propose or apply durable fixes.

Frequently Asked Questions about retrospect

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

FAQPage Schema
How do I analyze agent session failures and prevent recurring mistakes?

Agent session failures are analyzed by scanning conversations for corrections and tool failures, classifying root causes, and converting verified problems into durable instruction patches and enforcement mechanisms to prevent recurrence.

What is root cause analysis for AI agent behavioral problems?

Root cause analysis for agent behavior classifies session failures across missing, undiscovered, ignored, ambiguous, or incorrect context to identify why mistakes occurred and prescribe targeted fixes.

How do I create instruction patches from conversation analysis?

Instruction patches are created by diagnosing conversation alignment gaps, verifying whether existing enforcement caught the failures, and prescribing fixes across rule cards, stop classifiers, tool gates, and skills to prevent recurrence.

Can I automate retrospective workflows after autopilot or deep-qa coding sessions?

Retrospective workflows can be automated after autopilot or deep-qa sessions by analyzing what went wrong, verifying existing enforcement mechanisms, and producing targeted patches with a concise summary.

Does root cause analysis require access to hook logs and repository workflows?

Root cause analysis requires access to the current conversation, relevant configuration, hook logs, rule cards, skills, and repository workflows to accurately diagnose failures and apply durable fixes.

What's the best way to fix repeated agent corrections during coding tasks?

Repeated agent corrections are fixed by scanning the session for behavioral signals, classifying the root cause of the corrections, and applying durable enforcement patches across pre-commit hooks and documentation.