skills.rest
hpark0011
hpark0011COMMUNITY
Explore all skills by this author →
GitHub

retro

Analyzes Claude Code session transcripts and scores them against development process rules.

Updated Apr 17, 2026
Search Tags:#claude code#retrospective#process improvement#session analysis#jsonl#scoring rubric#transcript parsing
One-click install
npx skills add https://github.com/hpark0011/mirror --skill retro-hpark0011
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: retro
Source: https://github.com/hpark0011/mirror/tree/main/.agents/skills/retro
Command: npx skills add https://github.com/hpark0011/mirror --skill retro-hpark0011

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Reviewing past AI coding sessions manually is slow and subjective, making it hard to spot recurring anti-patterns like bandaid fixes, skipped investigation, or debugging spirals. This Skill automates retrospective analysis of JSONL session transcripts and produces a structured, evidence-based report. ## Core Features & Use Cases - Programmatic Extraction: Python scripts parse large JSONL session files into compact JSON summaries with tool counts, signal flags, and investigation ratios, avoiding context overflow. - Rubric-Based Scoring: Scores each session across 7 weighted dimensions (session discipline, problem-solving flow, fix quality, debugging efficiency, communication, context management, tool efficiency) with cited evidence. - Structured Reporting & Proposals: Outputs a formatted report with dimension trends, per-session breakdowns, and diff-style proposals for MEMORY.md and dev-process.md updates that require explicit user approval. - Use Case: Run /retro 7d at the end of a week to review all recent sessions, identify that investigation ratios are consistently below 1.0, and receive a proposed new dev-process rule requiring hypothesis statements before edits. ## Quick Start Ask the assistant to run /retro 5 to analyze the last five sessions and produce a scored retrospective report with proposed memory updates.

Frequently Asked Questions about retro

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

FAQPage Schema
How do I analyze past Claude Code sessions for process issues?▼

Invoke /retro with a scope argument such as /retro 5 for the last five sessions, /retro 7d for the last seven days, or /retro all. The skill extracts compact summaries from JSONL transcripts, scores them against a rubric, and outputs a structured report.

What does the session retrospective report include?▼

The report contains an executive summary, a weighted dimension scores table with trends, per-session breakdowns with evidence and flags, cross-session patterns, reusable learnings, and diff-style proposals for MEMORY.md and dev-process.md updates.

How are sessions scored in the retrospective?▼

Each session is scored 1-5 across seven weighted dimensions including session discipline, problem-solving flow, and clean fix quality. Scores must cite extraction signals like tool counts, revert keywords, and investigation ratios, then normalize to a 5.0 scale.

Can the retro skill handle very large session transcript files?▼

Yes. The Python extraction scripts read JSONL files line-by-line to avoid loading multi-megabyte files into memory, and produce compact 1-2KB JSON summaries. Reading raw JSONL directly into context is explicitly prohibited.

Does the retro skill automatically update MEMORY.md or dev-process.md?▼

No. All proposed updates are presented as diff blocks for user review and require explicit approval before any changes are written. The skill is limited to 3-5 high-value proposals to avoid noise.

Related Skills

Explore More Software Engineering Skills→
BinyaminEdenBinyaminEden
26

continuous-learning

Extract reusable coding patterns from Claude Code session transcripts.

Community
Intermediate
damilola-elegbede-orgdamilola-elegbede-org

verify

Assess executed commands against defined requirements and generate alignment scorecards.

Official
Intermediate
basicmachines-cobasicmachines-co
25

memory-reflect

Review recent notes and conversations to consolidate insights into MEMORY.md.

Official
Intermediate
silversteinsilverstein
1.4k

minutes-mirror

Analyze meeting transcripts for talk-time, filler, and hedging metrics.

Community
Advanced