EngineeringAudits your codebase overnight, one cleanup per area
Nightly auditor
Each night it studies the whole repository before touching anything, keeps a ledger of what it found split into what it knows and what it suspects, and picks at most one cleanup per area of the code. Every cleanup is a small pull request that deletes or reuses before it adds, fixes causes rather than adding guards, and comes with test output it actually ran. In the morning you get a short report and a few pull requests to merge or close.
- Works with
- GitHubSlack
- Recommended model
- Claude Opus 5.5
- Skills
- 7
What it does
- Maps the repository into areas and reads all of it before proposing any change
- Keeps a findings ledger that separates what it proved from what it only suspects
- Opens at most one small cleanup pull request per area each night, smallest diff first
- Prefers deleting and reusing code over adding guards, flags or new machinery
- Runs your tests and linter on its own computer and attaches the real output
- Leaves a short morning report and never merges
The skills it brings
Playbooks the mate follows when the job calls for them. You can read and edit every one after adding it.
/nau-setupnau-setupUse when setting up the nightly audit for the first time, adding a repository, or changing when and how the audit runs.
/nau-surveynau-surveyUse when auditing the codebase, before any cleanup, to read every file in the chosen areas, update the area map and record findings with evidence in the ledger.
/nau-cleanup-prnau-cleanup-prUse when turning one ledger finding into a small, verified cleanup pull request, at most one per area per night.
/nau-morning-reportnau-morning-reportUse when the nightly run finishes, or the person asks what the audit found, to write a short report of pull requests opened, findings logged and anything that needs their decision.
/eng-repo-setupWorking copy setupGetting a working copy of the person's code on your computer before any engineering task: access (GitHub connector, a token they paste, or a Knowledge folder), cloning, a branch of your own, installing dependencies and a baseline test run.
/eng-systematic-debuggingSystematic debuggingFixing a bug, failing test, broken build or unexpected behaviour in the person's code: find the root cause with evidence before changing anything, prove it with a failing test, fix once, verify. Use before proposing any fix.
/eng-verificationVerify before saying doneAbout to say code work is done, fixed, passing or ready (before a commit, a pull request, or a status update to the person): run the command that proves it, read the output, and report what it showed. Evidence before claims.
Things to ask it
Set up a nightly audit on …Run an audit now and show me the findingsWhat did you find last night?Skip this area from now on: …