research-repo-technology
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources like GitHub repositories and websites during technology research, creating a surface for instructions embedded in that content to influence the agent's findings or subagent instructions. 1. Ingestion points: The skill searches and reads external source files, project issues, and documentation as specified in SKILL.md. 2. Boundary markers: The skill establishes a 'research boundary' in SKILL.md that prohibits repository mutation and requires tasks to remain read-only. 3. Capability inventory: The agent is authorized to browse the web and spawn subagents with high reasoning capabilities. 4. Sanitization: No specific technical sanitization or filtering is defined for external content, relying instead on model reasoning.
- [DATA_EXFILTRATION]: The skill's instructions require the agent to access the 'live repository truth' (including source code and configuration manifests) while simultaneously performing network operations to browse for external facts, creating a potential data exfiltration surface.
Audit Metadata