rithvik-duddupudi/Eris LLM Security Scan
Automated LLM security testing with Red Team attacks. Gate deployments based on vulnerability scores.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| eris_api_url | URL to the Eris API (e.g., https://chaosml-api.onrender.com) | yes | — |
| eris_api_key | Eris API key for authentication | yes | — |
| llm_provider | LLM provider to test: openai, gemini, anthropic | yes | — |
| llm_api_key | API key for the LLM provider being tested | yes | — |
| llm_model | Model name to test (e.g., gpt-4, gemini-pro) | no | "" |
| vulnerability_threshold | Maximum acceptable vulnerability score (0-100). Workflow fails if above this. | no | 30 |
| attack_categories | Comma-separated attack categories: jailbreak,prompt_injection,data_exfiltration,policy_bypass,manipulation,hallucination | no | "" |
| fail_on_critical | Fail immediately if any critical-severity attack succeeds | no | true |
Outputs
| name | description |
|---|---|
| vulnerability_score | Vulnerability score from the security scan (0-100, lower is better) |
| risk_rating | Risk rating: Critical, High, Medium, Low |
| passed | Whether the test passed the threshold (true/false) |
| successful_attacks | Number of attacks that succeeded (vulnerabilities found) |
| total_attacks | Total number of attacks run |
| report_json | Full scan results as JSON |