receipts
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses strong imperative language to redefine the agent's reasoning style and identity, instructing the agent to 'drop' performed insight and avoid agency-themed language, which overrides default operational behaviors.\n- [COMMAND_EXECUTION]: The skill includes a shell command snippet used to probe the local filesystem for the existence of overlay configuration directories.\n- [DYNAMIC_EXECUTION]: The agent is instructed to dynamically determine file paths for additional instructions at runtime based on the value of the '$POWERSUIT_OVERLAY' environment variable.\n- [INDIRECT_PROMPT_INJECTION]: The skill implements an overlay mechanism that allows external, potentially untrusted files to override internal logic without sanitization or boundary markers.\n
- Ingestion points: Files located within the directory specified by '$POWERSUIT_OVERLAY/stance/'.\n
- Boundary markers: None; the skill explicitly states that the agent should 'follow the overlay' if it disagrees with the base instructions.\n
- Capability inventory: The agent is directed to read external files and prioritize their instructions over its default stance.\n
- Sanitization: No validation or filtering of the external overlay content is performed.
Audit Metadata