lov-dev-research
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCECOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill researches technology by ingesting external data from official documentation and project source files, creating a potential surface for indirect prompt injection where external content could attempt to influence agent behavior.
- Ingestion points: Web research results, official documentation, and project source files (SKILL.md).
- Boundary markers: The 'Execution boundary' section in SKILL.md establishes operational rules and manual review steps, though specific prompt delimiters for external content are not defined.
- Capability inventory: The skill utilizes the host agent's file system, search, and CLI tools (SKILL.md).
- Sanitization: Instructions require verification of facts against official documentation (Workflow step 3), providing a manual sanity check against potentially injected misinformation.\n- [PERSISTENCE]: The skill uses a script
scripts/profile_store.pyto persist user-owned profile context in local JSON files (e.g., in~/.config/agent-skills/). This mechanism allows the agent to maintain state across different sessions for identity and brand consistency. The script includes a blacklist for sensitive keywords to prevent storing credentials and uses atomic writes with appropriate file permissions.\n- [COMMAND_EXECUTION]: The skill is designed to execute local Python scripts (scripts/profile_store.pyandscripts/validate_skill.py) for maintaining its profile database and performing source validation. These executions are intended for skill administration and operate on local files provided within the skill package.
Audit Metadata