product-tracking-model-product
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill performs a 'Silent Codebase Scan' that ingests content from external files like README.md, package.json, and database schemas. This creates a surface where malicious instructions embedded in a codebase could influence the agent's behavior.
- Ingestion points: External project documentation, routing files, API controllers, and models scanned in SKILL.md.
- Boundary markers: None; the skill does not specify delimiters or instructions to ignore embedded prompts in the scanned files.
- Capability inventory: The skill is capable of writing telemetry artifacts to the file system and executing shell commands (mkdir).
- Sanitization: Not present; the agent is instructed to synthesize and summarize the data without explicit sanitization or filtering logic.
- [COMMAND_EXECUTION]: The skill instructs the agent to run shell commands (
mkdir -p .telemetry/audits) during its initialization phase. These are benign and limited to local directory management within the skill's scope.
Audit Metadata