writing-technical-design
Pass
Audited by Gen Agent Trust Hub on Apr 19, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill instructs the agent to use
WebSearchandWebFetchto research technical components from external online sources as part of its core workflow. - Ingestion points: Web search results and content from fetched documentation pages, repositories, and blogs (referenced in
DEEP-RESEARCH.md). - Boundary markers: Absent; the instructions do not specify the use of delimiters or 'ignore-instruction' blocks when processing retrieved data.
- Capability inventory: File system write access, web tool usage, and the ability to spawn subagents for parallel research tasks.
- Sanitization: Absent; the skill processes research findings directly into architecture documents and new skill files without explicit validation or filtering of embedded instructions.
- [COMMAND_EXECUTION]: Dynamic Instruction Generation and Persistence. The skill generates new instruction files under the
skills/tech-{component}/directory. These files are designed to be automatically discovered and loaded by the agent platform during future implementation sessions. This creates a mechanism where malicious instructions ingested during the research phase could be converted into persistent, trusted project instructions, leading to potential long-term behavioral overrides.
Audit Metadata