lambdalabsml/AgentBeats Competition Submission
Submit your attacker or defender solution to the AgentBeats Competition 2026
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Feb 24, 2026
- License
- None
Pinned Snippet
uses: lambdalabsml/agentbeats-submission-action@bef3e4100b6641bd45b8a0005793a9f628463c69 # no releases — HEAD as of 2026-07-11tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| api_key | API key for the competition (store in GitHub Secrets as COMPETITION_API_KEY) | yes | — |
| submission_endpoint | AWS Lambda endpoint URL for submissions | no | https://r6c2lb8bwl.execute-api.us-east-1.amazonaws.com/submit-solution |
| role | Submission role: "attacker" or "defender" | yes | — |
| submission_path | Path to the directory to submit (will be zipped) | no | ./src |
| run_tests | Integration test mode: 'off' or 'false': Skip testing 'warn' or 'true': Run tests, show warning if failed, but still submit (default) 'required' or 'block': Run tests, block submission if failed | no | true |
| openai_api_key | OpenAI API key for running tests (required if run_tests is true) | no | — |
| openai_base_url | OpenAI base URL for running tests (required if run_tests is true) | no | — |
| print_info | Whether to print submission info after upload | no | true |
| scenario | Scenario to test against: "thingularity", "portfolioiq", "medical", "gymjailbreak", "resume_downgrade", or "all" | no | resume_downgrade |
| test_only | If true, only run tests without submitting (useful for parallel test→submit workflows) | no | false |
Outputs
| name | description |
|---|---|
| submission_number | Total submission number for this team |
| role_submission_number | Submission number for this specific role |
| attacker_submissions | Total attacker submissions |
| defender_submissions | Total defender submissions |
| message | Response message from the server |
| test_passed | Whether integration tests passed (true/false/skipped) |