co-maintainer

probe

Read-only research on a GitHub repository. probe samples pull requests and commit history, then prints a recommended co-maintainer init command with sensible --include-* and --max-* flags, plus an estimate of what that command will cost in time and tokens. It does not call the AI and does not write guide files.

Before you run

Check Why
Install co-maintainer probe is a CLI command
GitHub auth (gh auth login or a PAT) probe only reads the repo via the API
Defaults with co-maintainer set (optional) Skips repeating --auth=... on every run
Read access to owner/repo Private repos need a token or gh login that can see them

If probe fails, fix GitHub access first (gh auth status), not co-maintainer.

Usage

co-maintainer set --auth=gh
co-maintainer probe owner/repo
co-maintainer probe              # repo from the git remote of the current directory
co-maintainer probe owner/repo --run   # probe, then run the recommended init

Output is a recommended block: a full co-maintainer init ... line you can copy, followed by an estimate and a next block.

Parameters

Flag Meaning
owner/repo Repository to research (optional, read from the git remote when omitted)
--auth=gh Use the GitHub CLI (default when unset: gh, or value from set / config)
--auth=pat Use GITHUB_TOKEN, GH_TOKEN, --github-pat=..., or set --github-pat=...
--github-pat=... PAT for --auth=pat when not in env or config
--env=PATH Load env vars from a file (see Configuration)
--gh-concurrent=N Parallel GitHub requests while sampling PR details (default 1, minimum 1)
--run After printing the plan, run the recommended init in the same process
--debug Verbose logs on stderr
--log-time Print timing for each probe phase

probe ignores --include-*, --max-*, and AI flags. Those appear on the recommended init command in the output when the analysis suggests them.

The estimate

The estimate block says how many AI jobs the recommended init would queue (extract and synth), a token range, a time range, and a dollar range. It is a planning estimate, not a bill:

  • Job counts come from the same inputs the real run uses, so the shape is exact.
  • Dollars need a price for both --low-model and --high-model. They come from OpenRouter's public /api/v1/models, cached for 24 hours. With no price there is no dollar line rather than a guessed one.
  • Time and per-job tokens come from this repository's own recorded jobs when four or more exist, and from the cm-dx-lab calibration otherwise. The line says which one it used.

--run runs the recommended init in the same process after printing the plan, which is the "look then do" step in one command. It is the same code path the dashboard uses, not a re-spawn.

What it looks at

  • Repository metadata and latest release (if any)
  • Pull request list (updated time, sampling across years)
  • A bounded sample of PR diffs for size and activity
  • Default-branch commit history

Use the printed init line as-is for a first run, or tighten limits if the repo is huge and you want a cheaper init. Next step: init.