Map agent authority in CI without pretending every match is a vulnerability.
FOXIFY Agent Authority Check is a dependency-free GitHub Action for MCP servers and tool-using AI agents. It inventories authority-bearing code paths and produces evidence JSON plus a deterministic evidence hash. It does not assign invented severity to a regex hit or execute the target repository.
Drop it into a workflow
name: Agent Authority Check
on:
pull_request:
push:
branches: [main]
jobs:
authority-preflight:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: speeedy10/agent-authority-check@v1
with:
path: .
output: foxify-authority.json
- uses: actions/upload-artifact@v4
with:
name: foxify-authority-evidence
path: foxify-authority.json
What the Action reports
- Tool registration and exposed action surfaces.
- Credential and ambient-secret access.
- Network, filesystem, shell/process, database and browser authority.
- Payment and transaction-related code surfaces.
- File/line observations, review-class counts and evidence SHA-256.
The evidence boundary
The Action deliberately returns OBSERVATIONS_ONLY. A source match does not prove a vulnerability. FOXIFY promotes a security finding only when the primitive, reachability and consequence chain is established.
Why teams put this in CI
Agent authority can change quietly when a pull request adds a new tool, widens credential scope, introduces shell or browser automation, or connects code to a payment or deployment surface. The Action gives reviewers a stable authority-oriented diff signal before that code reaches production, without executing the repository or requiring secrets.
The evidence artifact can also be retained with the build so a later review can compare what authority-bearing surfaces existed at a specific commit. That makes it useful as a preflight and audit input rather than a substitute for human security judgment.
From free CI to a bounded answer
Open the GitHub Action and source →
See the synthetic Quick Check report →
Run a 99-Star Quick Check for one public GitHub repository →
For production or explicitly authorized agent boundaries, use the human Authority Review or the runtime action preflight.