elliot-sones/AgentGate Security & Trust Scan
Run AgentGate security and/or trust scans on an AI agent as part of your CI/CD pipeline. Fails the step when findings exceed the configured threshold.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| scan-type | Which scan to run. One of: "trust", "security", or "both". Use "trust" to run static + runtime trust checks, "security" for red-team probing, or "both" to run the full combined scan. | no | trust |
| source-dir | Path to the agent source directory (relative to the repo root). Used for static trust checks and manifest discovery. | no | . |
| manifest | Path to the trust manifest file (relative to the repo root). Defaults to <source-dir>/trust_manifest.yaml. | no | trust_manifest.yaml |
| fail-on | Trust verdict threshold that causes the step to exit with a non-zero code. One of: "allow_with_warnings", "manual_review", or "block". | no | manual_review |
| agent-url | URL of a live hosted agent for runtime checks or security scanning. Required when scan-type is "security". Optional for trust scans. | no | "" |
| output-dir | Directory where scan reports are written. | no | agentgate-reports |
| agentgate-version | Version of agentgate to install. Defaults to the latest release. | no | "" |
| anthropic-api-key | Anthropic API key used by AgentGate for LLM-based analysis and adaptive attacks. Pass this via a repository secret. | no | "" |
Outputs
| name | description |
|---|---|
| verdict | Trust verdict string from the scan (e.g. "allow", "allow_with_warnings", "manual_review", "block"). For security scans this is the pass rate. |
| findings-count | Number of findings returned by the scan. |
| report-path | Path to the primary JSON report file. |