agent-guard

CI/CD safety and risk analysis for AI-assisted coding workflows (Claude Code, Codex, Cursor, Copilot, …).

agent-guard scans pull request diffs for the patterns that AI coding assistants tend to slip into your repo: hallucinated imports, dangerous calls, missing tests, infra/auth changes, and leaked secrets. It runs deterministically (no LLM required), in <10s on typical diffs, and posts a single sticky comment on your PR.

Status

Milestone 1 — MVP. Python only for hallucinated-import detection; all other analyzers are language-agnostic. See features/agent-guard-feature-plan.md for the full roadmap.

For the full reference — every CLI flag, every rule, every config key, CI recipes, troubleshooting, and FAQ — see docs/USER_MANUAL.md.

Quickstart

As a GitHub Action

# .github/workflows/agent-guard.yml
name: agent-guard
on: pull_request
permissions:
  contents: read
  pull-requests: write
jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: tasnuvaleeya/agent-guard@v0.1

Locally

pip install ag-scan                # PyPI distribution name
git diff main...HEAD | agent-guard scan --format md

Note: the PyPI distribution name is ag-scan because agent-guard was already taken by an unrelated project. The CLI command and GitHub Action are still named agent-guard.

What it flags

Each finding gets a severity (low/medium/high/critical); the report includes an aggregate risk score (0–100).

Configuration

Drop a .agent-guard.yml in your repo root. All keys are optional.

analyzers:
  secrets: { enabled: true }
  missing_tests: { enabled: true, min_added_lines: 30 }
fail_above: 60
exclude:
  - "vendor/**"
  - "**/*.generated.py"

Development

pip install -e ".[dev]"
pytest
ruff check .
mypy src

License

Apache-2.0