Review the local subagent-failure retro ledger, user-level config drift, and which consumer repos are behind the toolkit version. Triggers on /retro-review.
Scanned 9/13/2026
Install to Claude Code
npx -y skills add dagonet/claude-code-toolkit --skill retro-review --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Retro Review?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/dagonet-retro-review)More formats (shields.io, HTML) on the badges page.
---
name: retro-review
description: Review the local subagent-failure retro ledger, user-level config drift, and which consumer repos are behind the toolkit version. Triggers on /retro-review.
disable-model-invocation: true
argument-hint: "[project-dir]"
---
# Retro Review
The **local** counterpart to the `toolkit-nightly-check` routine. A cloud routine runs
over a *clone* and cannot see `~/.claude`, this machine's other checkouts, or anything
uncommitted — which is exactly what this skill reads. That is why it is a skill you run
here, not a routine.
Target project: `$ARGUMENTS`, or the current working directory if empty.
## Steps
1. **Read the retro ledger.** Slug the absolute project path exactly as
`hooks/retro-ledger.sh` does — replace every `:`, `\`, `/`, `.` and `_` with `-` —
then read `~/.claude/projects/<slug>/memory/retro.md`. No file means no subagent has
failed there; say so and continue. Group the entries by cause (a wrong `tools:`
allowlist, an under-briefed spawn prompt, a hook), not by date.
2. **Check user-level drift.** From the `claude-code-toolkit` checkout, run
`bash scripts/verify-user-level-drift.sh`. It compares `user-level-reference/` with
the live `~/.claude/`; the reference leads and the live copy follows, so DRIFT means
"sync is pending", not "broken". List the drifted files.
3. **Check consumer repos.** For each `G:/git/*/.claude/template-manifest.json`, read
its recorded template version and compare with line 1 of `VERSION` in the toolkit
checkout. List only repos strictly behind it, worst first.
4. **Write the summary** to `~/.claude/projects/<slug>/memory/retro-summary.md`:
**at most 30 lines**, overwriting any previous run. One section per step, each a
short list of concrete next actions ("re-copy `agents/architect.md`", "add `Skill`
to `tester.md` tools:"). Drop any section that found nothing. If all three are clean,
write a single line saying so.
## Running it unattended
```bash
claude -p "/retro-review" --permission-mode auto # from the project directory
```
Schedule that with Windows Task Scheduler or cron. It is a **local** headless session,
not a cloud routine: it needs this machine's `~/.claude` and its sibling checkouts, so
it cannot be moved to `/schedule`. Keep it read-only — this skill reports, and the fixes
are a separate, deliberate act.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!