recipe-validate
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from hypothesis files, user-supplied text, and product/design sources which could contain malicious instructions designed to alter the agent's logic.\n
- Ingestion points: Target hypothesis files (SKILL.md Step 1) and decision-relevant product/design sources (SKILL.md Step 3).\n
- Boundary markers: The instructions do not define specific delimiters or escaping mechanisms for the ingested content, though the architecture uses sub-agents for logical isolation.\n
- Capability inventory: Access to web search, filesystem writes (prototype generation), and code execution tools.\n
- Sanitization: No explicit logic or instructions are provided to sanitize or filter external content before it is processed by the agent or sub-agents.\n- [COMMAND_EXECUTION]: The skill utilizes a 'code execution' tool to perform 'code spikes' and dependency analysis during feasibility validation. This provides a high-privilege environment that could be misused if the agent's instructions are subverted through prompt injection.\n- [DYNAMIC_EXECUTION]: The 'prototype-generator' sub-agent dynamically creates self-contained HTML files in the
docs/discovery/prototypes/directory. If the input sources are compromised, this mechanism could be used to generate malicious artifacts or scripts within the project environment.
Audit Metadata