Local CLI

The demur CLI runs the same pipeline, profiles, lenses, finding contract, severity taxonomy, and budget machinery as the GitHub Action, reading the same .demur.toml from the repository root. It is the fastest loop for tuning profiles, lenses, and prompts, because a local review never pushes to a pull request.

Review local changes

# The working copy, including uncommitted changes
demur review

# A revision range
demur review main..HEAD

History is read through jj when the repository is jj-based and through git otherwise. Both produce the same review for the same diff. No GitHub API call is made for a local review.

Review a pull request

demur pr 42
demur pr https://github.com/owner/repo/pull/42

The target may be a pull request number inside a repository checkout or a full URL. The GitHub token comes from GITHUB_TOKEN, then GH_TOKEN, and finally from the GitHub CLI when it is authenticated (gh auth login). The pull request diff is fetched and reviewed in full: the CLI keeps no state, so it never reads or writes the bot's review markers.

Progress logs

Progress goes to stderr while a review runs: which pass is executing, how long it took, what it cost, and any degradation that applied. The default verbosity is info; set RUST_LOG=debug for more detail or RUST_LOG=error for near silence. Stdout stays clean, so piping the markdown output works unchanged.

Output formats and exit status

--format markdown (the default) prints the review ranked exactly as the pipeline ranked it. --format json prints one machine-readable document carrying the findings, severities, locations, verdict, and spend.

The exit status encodes the verdict so the CLI can gate a local workflow:

Status	Meaning
0	APPROVE
1	REQUEST_CHANGES
2	The run failed to complete

A failed run never reports as a clean review.

Publishing

The CLI never writes to GitHub by default. demur pr --publish posts the review, and it says so first: the findings will appear under the identity that owns the token, not under a bot identity. A locally published review carries no bot state, so it does not affect delta reviews.

Budgets apply locally

Budget estimation, the cap, and the degradation ladder apply to local runs exactly as they do to Action runs, and the printed review reports spend and any degradation that applied.