stage-launch
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python scripts to process build sheets and validate generated payloads. It uses standard library modules to manage file permissions and directory structures for the generated artifacts.
- [DYNAMIC_EXECUTION]: The skill generates a bash script intended for manual execution by the user. This script uses curl to interact with the Anthropic API and manages local state files to allow for resumable deployment of the agent environment.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided configuration data from a JSON build sheet. This data is interpolated into the final API payloads, representing a potential surface for instructions embedded in the input data to affect the behavior of the deployed agent.
- Ingestion points: The build-sheet.json file read by scripts/payload_generator.py.
- Boundary markers: No specific delimiters or safety instructions are added around interpolated fields.
- Capability inventory: The generated artifacts allow for network communication via curl and filesystem access for state management.
- Sanitization: Relies on standard JSON parsing without additional content-based filtering.
Audit Metadata