podium-rate-limit-survival
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including Podium API response headers, webhook event data, and trace files in CSV format.
- Ingestion points: API response headers (Retry-After) and CSV files processed by
scripts/bucket_simulator.pyandscripts/burst_smoother.py. - Capability inventory: The skill possesses network capabilities (
Bash(curl:*)), script execution (Bash(python3:*)), and file modification permissions (Write,Edit). - Boundary markers: The implementation includes defensive parsing logic for headers and explicit type casting for CSV fields, which limits the potential for control character injection.
- Sanitization: External data is converted into specific data types (floats, integers, ISO dates) before being used in logic or displayed to the agent.
- [COMMAND_EXECUTION]: The skill uses
Bashto execute local Python scripts and interact with Redis. These actions are aligned with the skill's purpose for rate-limit simulation and quota monitoring. - [EXTERNAL_DOWNLOADS]: The skill references standard, well-known libraries such as
httpx,redis, andaiolimiter. These dependencies are part of the common ecosystem for the described tasks.
Audit Metadata