Claude Retro Skill

claude-retro-skill

A `/retro` skill for [Claude Code](https://claude.com/claude-code): an end-of-session retrospective the agent runs on itself, with a human approval gate on every change to its own standing instructions and automation.

**In plain terms:** your AI assistant reviews its own work at the end of each session, suggests what it should learn, and you approve anything that would rewrite its own instructions.

Install

Two ways to install: run the commands yourself, or paste the prompt below and Claude does it for you.

Copy the `retro` folder into your Claude Code skills directory:

git clone https://github.com/robertantolin/claude-retro-skill.git
cp -r claude-retro-skill/retro ~/.claude/skills/retro

Use `~/.claude/skills/` to make it available in every project, or a project's `.claude/skills/` to scope it to that project.

**Or ask Claude Code to install it.** Paste this into a session and the agent handles the paths and platform differences:

Clone https://github.com/robertantolin/claude-retro-skill and copy its retro/
folder verbatim into my user-level Claude Code skills directory
(~/.claude/skills/retro on Mac/Linux, %USERPROFILE%\.claude\skills\retro on
Windows). Do not modify SKILL.md. Confirm the file exists, then tell me how
to invoke the skill.

Either way, end any working session with:

/retro

What ships in the folder

• SKILL.md: the retro itself.
• health.py: prints your store sizes, the Machine gotchas word count, and the age of the last scan. The retro calls it; you can run it any time: python ~/.claude/skills/retro/health.py.
• scan.py: an independent observer. Run it weekly: python ~/.claude/skills/retro/scan.py. It reads your Claude Code session transcripts, finds the corrections you gave the assistant, checks each against your recorded lessons, and writes ~/.claude/retro/findings.tsv. The next retro must address every open finding. It also records how you replied to past retros in `~/.cla