lov-agent-instructions
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes project instruction files like AGENTS.md and CLAUDE.md which could contain untrusted data. * Ingestion points: Reads AGENTS.md, CLAUDE.md, and project structure as documented in SKILL.md. * Boundary markers: Explicitly instructs the agent not to treat external repository text as system instructions. * Capability inventory: Uses host-provided file read, write, and command execution capabilities for auditing and improving instructions. * Sanitization: Implements project-specific versioning and convention preservation for all document changes.
- [PERSISTENCE]: The skill uses scripts/profile_store.py to maintain user preferences and skill records in a local profile.json file. This is a legitimate mechanism for cross-session context. The script proactively filters sensitive keywords (e.g., secret, password, api_key) to prevent accidental credential exposure and sets strict file permissions (0600).
- [COMMAND_EXECUTION]: The skill executes local commands for validation (scripts/validate_skill.py) and profile management. These are scoped to the skill's environment and intended for project maintenance.
Audit Metadata