shortform-production
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the execution of a Python-based API client (scripts/metricool.py) to manage video scheduling and delivery. It also references optional telemetry and version check scripts (telemetry/version_check.py, telemetry/telemetry_init.py) for maintenance purposes.
- [EXTERNAL_DOWNLOADS]: The skill interacts with the Metricool API (app.metricool.com) and AWS S3 storage for video delivery. These interactions target well-known, legitimate services and are core to the skill's functionality.
- [DATA_EXFILTRATION]: The skill transmits media content and API tokens to the Metricool service. The implementation follows security best practices, ensuring API tokens are handled via environment variables, not logged, and not forwarded to third-party storage providers.
- [INDIRECT_PROMPT_INJECTION]: The skill uses an LLM-based editorial judge prompt (references/eval/judge-prompt.md) which processes external video transcripts and source content. 1. Ingestion points: Video transcripts and source recording data are processed in the judge-prompt.md instructions. 2. Boundary markers: The prompt uses explicit procedural steps and calibration rules to guide the evaluation logic. 3. Capability inventory: The skill allows file writing (receipts) and network operations (Metricool API) via the API client script. 4. Sanitization: Input data is validated using JSON schema and path checks within the API client script.
- [DYNAMIC_EXECUTION]: The unit tests (tests/test_metricool.py) utilize standard Python dynamic loading (importlib) to import the API client script for validation purposes.
Audit Metadata