session-retro

Extracts learnings, skill candidates, and docs gaps from Claude Code session transcripts.

5|1|Updated Aug 18, 2023
One-click install
npx skills add https://github.com/thpoll83/PolyKybdHost --skill session-retro-thpoll83
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: session-retro
Source: https://github.com/thpoll83/PolyKybdHost/tree/main/.claude/skills/session-retro
Command: npx skills add https://github.com/thpoll83/PolyKybdHost --skill session-retro-thpoll83

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Valuable discoveries made during a long coding session — root causes, toolchain quirks, hard-won commands, and repeatable workflows — are easily lost when the session ends. This Skill scans the session transcript and turns those findings into durable CLAUDE.md notes, new reusable skills, and docs updates. ## Core Features & Use Cases - Learning extraction: Identifies non-obvious facts (root causes, environment quirks, exact commands) and drafts edits for the most specific CLAUDE.md or docs file. - Skill codification: Detects multi-step workflows repeated during the session and scaffolds new SKILL.md files with trigger-rich descriptions. - Docs gap and PR sweep: Flags user-facing changes missing from public docs and checks every PR touched for red CI and unread review feedback. - Use Case: At the end of a debugging-heavy session, ask for a retro to get a ranked proposal of what to record, then have the approved items written, committed, and opened as pull requests automatically. ## Quick Start Ask the assistant to run a session retro and scan this session for learnings and workflows worth keeping.

Frequently Asked Questions about session-retro

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

FAQPage Schema
How do I capture learnings from a Claude Code session?

Run a session retro at the end of the session. It scans the transcript for root causes, toolchain quirks, and exact commands, then proposes targeted edits to the most specific CLAUDE.md or docs file for your approval.

When should a workflow become a reusable skill?

Codify a workflow only if it is multi-step, repeatable on different inputs, has a clear trigger, and is not already covered by an existing skill. One-off tasks and single commands should stay as notes or aliases instead.

Where are Claude Code session transcripts stored?

Claude Code stores one JSONL file per session under the projects directory, such as /root/.claude/projects or ~/.claude/projects. Each line is a JSON event with types like user, assistant, tool_use, and tool_result.

Does the session retro write files without approval?

No. It first presents a ranked proposal of learnings, skill candidates, and docs gaps, then stops. Only after you approve specific items does it write the files, commit, push, and open pull requests.

What are the limitations of automated session retrospectives?

The retro depends on what is visible in the transcript, so context lost to summarization may be missed unless the full JSONL is read. It can also over-propose, so conservative judgment is needed to avoid codifying one-off work.