Commands

This page is a reference for the Argus command-line commands. Most Argus users never touch these — but if you’re contributing to the project or releasing new versions, this is your checklist.

A note on make: make is a common build tool that runs commands defined in a Makefile. You only need it if you’re working inside the Argus repository itself.

The short version

Daily commands

These are safe to run any time from the repository root:

make check-version    # Show current version + release status — e.g. "Current version: 0.6.0"
make clean            # Remove generated files (the dist/ folder)

make check-version is useful right before a release — it also reports whether the current VERSION already has a release tag (and make release would refuse a duplicate).

Quality checks

Run these after you edit review rules or documentation:

make validate         # Check SKILL.md trigger phrases, CHANGELOG, required files, Python syntax
make test-fixtures    # Run fixture regression tests (static mode — no API key needed)
make eval-gate        # Enforce the fixture regression lock vs config/quality-baseline.json
make test             # Everything: validate + test-fixtures (the full pre-release check)

Why: validate makes sure nothing is broken (missing files, inconsistent version, malformed scripts). test-fixtures confirms the review rules still produce the expected results on sample files. make test is both together — run it before any release.

Quality is measured on two levels — don’t confuse them:

• make eval / make eval-gate score the deterministic static scanner against the fixture expectations. This is a fixture regression lock (it catches scanner + fixture drift); it is not a measurement of the LLM agent.
• make eval-llm runs the quality engine in LLM mode (--mode llm, pinned model) and is the agent-quality signal. It is tracked over time (config/quality-llm-history.jsonl), not hard-gated, because free-model output varies run to run.

Version bumping

These update the version number and the changelog for you:

make bump-patch   # e.g. 0.6.0 → 0.6.1 (bug fixes, minor changes)
make bump-minor   # e.g. 0.6.0 → 0.7.0 (new features)
make bump-major   # e.g. 0.6.0 → 1.0.0 (breaking changes)

Why it matters: bump targets automatically update VERSION and CHANGELOG.md and stage the changes. You then review, run make test, and release.

Releasing

make test        # validate + fixture tests (full pre-release check)
make release     # release-gate → verify → tag → push

Why it matters: make release refuses to re-release an existing tag or an older version, verifies the version files are committed, creates the v*.*.* tag, and pushes it. Pushing a v*.*.* tag triggers the release workflow, which validates versioning, builds the release archives, and publishes a GitHub Release. A daily Release Check workflow additionally fails if VERSION is ever bumped without a matching tag — the v0.5.0 incident (files said 0.5.0, no release or skill package ever existed).

Only maintainers should run make release — it pushes to the shared repository.

Using Argus (not developing it)

You don’t need make at all to use Argus:

• GitHub App: install argus-flash and add a workflow — see Getting Started. GitHub runs everything for you.
• Local agent: point OpenCode, Claude Code, or Codex CLI at the repository. Argus loads automatically from AGENTS.md + SKILL.md; no commands required.

Advanced: running the tools directly

For contributors who want finer control:

make test-fixtures-llm runs the fixture suite against a real model, so it needs an OpenCode CLI with a configured model. Configure once with opencode auth login (choose OpenCode and paste your key from opencode.ai/auth), or set the OPENCODE_API_KEY environment variable.