agency-evidence-collector
Warn
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The agent is instructed to execute a local shell script named
./qa-playwright-capture.shto automate browser tasks. - [COMMAND_EXECUTION]: The skill requires the agent to run filesystem commands including
ls,grep, andcatto inspect project structure and content. - [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8) due to its ingestion of external test results and specifications without strict boundary markers or sanitization.
- Ingestion points: The skill reads
test-results.jsonand external specifications. - Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potential embedded commands in the ingested data.
- Capability inventory: The agent has access to shell execution and filesystem read/search commands.
- Sanitization: There is no documented validation or filtering of external data before it influences agent behavior.
Audit Metadata