CLI reference¶

Inside a workspace run commands as ./triage.py <command>; elsewhere use ai-triager --workspace DIR <command>. Every command has --help.

Setup¶

Command	What it does
`init [DIR]`	Set up a workspace (`--repo`, `--checkout`, `--python`, `--skills`, `--name`, `--package`)
`check`	Show what is configured and what is missing
`doctor`	Check gh, the project environment and the browser pipeline
`sandbox`	Probe the execution sandbox and the guard
`keys`	Show which providers have keys; `--set VAR` stores one
`skills`	List skills available to the built-in agent
`components`	Regenerate the component map
`app`	Serve the web app (`--show`, `--port`, `--dev`)

Command	What it does
`sync`	Fetch the open issue list (`--details`, `--stale`, `--issue N`)
`stale`	List writeups whose issue changed after triage
`context N`	Print an issue's body, comments, cross-references and hints
`next`	Claim the next issues and scaffold writeups (`--agent`, `--review`, `--issue`, `--label`, `--order`)
`run N`	Run `repros/N.py` with the guard, sandbox and a timeout
`browse N`	Serve `repros/N_app.py` and inspect it in headless Chromium
`validate [N...]`	Validate writeups
`finish N`	Stamp, validate, release the claim and reindex
`verify N --result yes\\|disputed`	Record a review verdict
`release N`	Drop claims
`index`	Regenerate `INDEX.md`

Command	What it does
`launch`	Start a background triage (or `--review`) job: `-m MODEL`, `-c COUNT`, `-j WORKERS`, `--runner`, `--issue`, `--label`, `--effort`, `--steps`
`jobs`	List jobs
`stop JOB`	Stop a running job

Command	What it does
`outbox list`	Show everything staged for GitHub, with warnings
`outbox send`	Send it; a dry run unless `--yes`, items with warnings only with `--force`
`log`	Every change ai-triager made on GitHub (`--limit`)
`undo ID`	Revert a change from the log

Command	What it does
`verdicts score`	Score writeups with the decision model (`--scope close\\|unscored\\|history\\|all`, `--issue`, `--background`)
`verdicts list`	Show scores, most likely closable first

Command	What it does
`calibrate`	Compare model confidence with your decisions
`calibrate --import-history`	First import closes from GitHub, `replies/*.md` verdicts and earlier decisions
`calibrate --save-thresholds`	Learn per-label thresholds (used when `[labels] learned_thresholds = true`)

Command	What it does
`schedule start` / `stop`	Run scheduled triage in the background, or stop it
`schedule status`	Whether it runs, the last and next run, the latest digest
`schedule once`	One pass: sync, triage new issues, write a digest

Command	What it does
`closing`	List issues recommended for closing and their decisions
`decide N accept\\|rereview\\|keep\\|clear`	Record a decision (`--note`)
`rereview [N...]`	Send re-review decisions back to a reviewer (`--launch`, `--model`)
`close [N...]`	Comment on and close accepted issues; a dry run unless `--yes`

Command	What it does
`labels sync`	Fetch the label catalogue and issue types
`labels classify`	Review labels (`--scope`, `--count`, `--model`, `--workers`, `--issue`, `--background`)
`labels list`	Show suggestions (`--all` includes applied and dismissed)
`labels apply --issue N`	Apply suggestions; a dry run unless `--yes`

Command	What it does
`prs sync`	Fetch open pull requests
`prs review [N...]`	Review PRs (`--scope unreviewed\\|external\\|all`, `--limit`, `--workers`, `--model`)
`prs list`	Show results with failed checks
`prs comment N`	Draft a comment for the contributor (`--model`)