Skip to main content
skill CX Codex

retro-meta — Claude Code skill: what it does, how to install

Runs a deeper investigation across past retros and the current ax setup to surface fixes the regular heuristic missed. Triggered by phrases like "deep retro" or the /ax:retro-meta command.

AGPL-3.0 ★ 124 updated 2026-10-07

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add Necmttn/ax --skill retro-meta -a claude-code
Open on GitHub

What retro-meta does

The skill hands reasoning to an external agent, Claude Code or Codex with a high thinking budget, while the ax CLI just gathers structured evidence and takes back any plan the user approves. It first checks already-accepted experiments that still lack a locked verdict, closing them with ax improve verdict or leaving them open when there isn't enough signal yet. It then works through the investigation prompts, reading skill files and the CLAUDE.md guidance files, and proposes findings for the user to confirm.

It needs axctl on PATH; if ax doctor fails, the skill stops and points to the DuckDB setup section in the docs. A useful run also needs at least three retros logged in the past month — fewer than that, and the skill suggests running ax:retro first to build up evidence.

It never writes a new skill file directly. Instead it drafts a plan under ~/.claude/plans/, asks for an explicit yes, and only then registers it through ax retro plan. Approved plans can optionally be handed to an internal scaffolding agent via ax improve accept --with-agent.

What it helps with

  • spot ax setup improvements the standard heuristic doesn't surface
  • resolve stalled experiment proposals and lock their verdicts
  • check skill files and CLAUDE.md for duplicates before proposing changes
  • draft a change plan backed by evidence from prior retros
  • hand an approved plan off for automatic skill or hook scaffolding

Example prompts for the agent

  • Run a deep retro on my ax setup
  • /ax:retro-meta
  • Investigate what's wrong with my agent setup

How to install retro-meta

1Skills CLI (Vercel)

npx skills add Necmttn/ax --skill retro-meta -a claude-code

2Manually: put the files in place

~/.claude/skills/retro-meta/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/retro-meta/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about retro-meta

How do I install retro-meta in Claude Code?

Run npx skills add Necmttn/ax --skill retro-meta -a claude-code; the skill lands at ~/.claude/skills/retro-meta/SKILL.md.

What does retro-meta do?

It runs a deeper pass over accumulated retros and the current ax setup to find improvements the regular heuristic proposals miss, leaving the final call on any new plan to the user.

What does retro-meta require to run?

It needs axctl on PATH and at least three retros logged within the past 30 days, or it recommends running ax:retro first.

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set12

The whole set →
tune tune reviews one coding session and records proposed agent-environment changes in ax. Use it to trace tool errors, wasted calls, and missing checks. CX Codex +1 ★ 124repo stars ship-checklist ship-checklist — a definition-of-done skill for the ax graph: it checks that a new write has a read path an agent can find, not just a manual CLI run. CX Codex +1 ★ 124repo stars setup setup installs the ax environment, builds a local DuckDB session graph, and validates the agent configuration. CX Codex +1 ★ 124repo stars retro retro runs a guided session over the ax agent-experience graph. Shows open proposals, pending verdicts, and hook effectiveness signals. CX Codex +1 ★ 124repo stars release-announcement Drafts detailed release announcements and changelog pages for the ax project. It pairs Git commit ranges with agent session logs to explain why code changed. CX Codex +1 ★ 124repo stars ax-narrate The ax-narrate skill turns an agent’s current-session context into a structured story, including code changes, user corrections, abandoned approaches, and consequential tool errors. CX Codex +1 ★ 124repo stars efficient-dispatch efficient-dispatch — a skill that keeps judgement work on the strong model, sends mechanical subagent tasks to a named cheaper model, and checks the routing against your own logs. CX Codex +1 ★ 124repo stars dojo dojo runs a background self-improvement loop using surplus model quota. It processes maintenance agendas in isolated worktrees via the ax tool. CX Codex +1 ★ 124repo stars

Project sponsors

Companies that keep this analytics open