stelow-product-job-to-be-done
Pass
Audited by Gen Agent Trust Hub on Jul 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The reference files describe the use of several external packages for tool fallbacks, including
@earendil-works/pi-agent-browser,cursor/plugins,Prinova/pi-agent-codebase-workflows, and@tintinweb/pi-tasks. These are documented as part of the universal fallback mechanism for different AI agent platforms. - [COMMAND_EXECUTION]: The skill's reference documentation includes several shell and Node.js snippets used for workflow management. Examples include
file-locking.md(atomic file operations vialn),read-config.md(parsing JSON vianode -e), andexecution-loop.md(managing implementation cycles via bash). - [DATA_EXFILTRATION]: The
agent_browser.mdtool allows for automated web browsing and visual inspection of URLs provided by the user. While this involves network access, it is a primary function of the tool for UI/UX auditing. - [PROMPT_INJECTION]: The skill uses a surface for indirect prompt injection by interpolating user-provided text (such as market definitions and job descriptions) into complex research prompts.
- Ingestion points: User-provided market definitions and situational variables enter the context in files like
07-functional-needs.mdand10-job-map-steps.md. - Boundary markers: Prompts use
<segment>and<situational factors and variables>tags to delimit injected data. - Capability inventory: The orchestrator environment described in the references includes
bash,subagentdelegation, andagent_browser(file-write/network access). - Sanitization: The prompts include specific instructions for the AI to 'abstract' or 'filter' inputs if they specify solutions or break formatting rules.
Audit Metadata