CLI Reference
drep v3.1.0 has six command groups. The binary's --help output is authoritative for the installed version.
Command overview
| Command | Purpose |
|---|---|
drep check | Run deterministic tools and LLM review on local paths or Git-selected changes |
drep lint-docs | Rule-based Markdown linting with no LLM or network |
drep doctor | Report detected languages, tools, and provider readiness |
drep init | Write drep.toml and optional Git hooks |
drep auth | Manage endpoint-keyed credentials without printing them |
drep acknowledge | Suppress a finding until its source context changes |
drep check
drep check [OPTIONS] [PATH]...| Option | Meaning |
|---|---|
--staged | Check files staged for commit |
--diff <REF> | Check changes since a base reference |
--tip <REF> | Choose the diff tip instead of HEAD; valid with --diff |
--pre-commit-push | Read the pushed base and tip from pre-commit's environment |
--format text|json | Select human or machine-readable output |
--fail-on info|warning|error | Also block on LLM findings at or above the chosen severity |
--cache-only | Use only cached LLM reviews; an uncached file exits 3 |
--push-gate | Warm cold reviews and request a fresh push connection |
--max-review-rounds <N> | Override the repository's maximum fresh LLM review rounds; N must be at least 1 |
--unlimited-reviews | Explicitly permit fresh LLM review rounds without a limit for this invocation |
drep check --staged
drep check --diff origin/main
drep check --max-review-rounds 5
drep check --unlimited-reviews
drep check --fail-on error
drep check --format jsonAuthoritative staged, diff, pre-commit-push, and bare push-gate checks allow three fresh, finding-producing semantic-review rounds per branch and worktree by default. At the limit, cached reviews and deterministic tools still run, while a cold semantic miss becomes an unanalyzed result and exits 2. JSON output's review object reports counted, reset, and explicitly unlimited states.
drep lint-docs
Lint Markdown without loading an LLM configuration or contacting the network.
drep lint-docs
drep lint-docs --staged --fail-on error
drep lint-docs --strict--strict is shorthand for --fail-on info. The published hook uses --fail-on error so it blocks rendering-breaking defects without turning whitespace into a commit failure.
drep doctor
drep doctor [PATH]
drep doctor --config alternate.tomlThe report includes machine-level site policy, applied provider concurrency ceilings, repository refusal, protocol-key status, and the effective request-header names for each HTTP provider without printing their values. If policy refuses review or cannot be evaluated, doctor skips the auth store, credential commands, and Codex probing.
drep init
Without --provider, an interactive terminal runs the wizard. Available presets are local, openrouter, zai, minimax, kimi, openai, codex, and custom.
drep init
drep init --provider codex --hooks both
drep init --provider custom --endpoint http://localhost:1234/v1 --model local-model
drep init --hooks none--path <DIR> selects a repository, --model <MODEL> overrides the preset model, and --non-interactive or --interactive explicitly controls whether the wizard may prompt. --force authorizes replacement of an existing drep.toml or a drep-managed hook. --no-gitignore keeps drep init from adding drep.toml to the ignore file.
drep auth
drep auth list
drep auth login --provider kimi
drep auth login --endpoint https://example.com/v1
drep auth logout --endpoint https://example.com/v1No auth command prints a stored key.
drep acknowledge
drep acknowledge <fingerprint> [<fingerprint>...]Fingerprints come from LLM findings. The acknowledgement remains valid only while the surrounding source context is unchanged.
Exit codes for check
| Code | Meaning |
|---|---|
0 | All required analysis ran; nothing blocks |
1 | Blocking findings |
2 | Required analysis did not run, including site-policy refusal or credential-command failure |
3 | Cache-only miss or successful push-gate warm requiring a fresh push |