lov-distill
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local Python scripts for configuration management and source validation.\n
scripts/profile_store.pyis invoked to read and atomically update user preferences in a local profile file (typically located at~/.config/agent-skills/profile.json).\nscripts/validate_skill.pyis used to ensure the skill source adheres to required structure and metadata standards.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided solution descriptions, creating a surface for potential injection.\n- Ingestion points: Solution descriptions and task evidence are extracted from the user's session or specified project files in
SKILL.md.\n - Boundary markers: The instructions explicitly separate analysis, preview, and modification phases, and define strict triggers for distillation tasks.\n
- Capability inventory: The skill performs local file system operations (writing Markdown documents and JSONL indices) and calls local helper scripts.\n
- Sanitization: The workflow mandates explicit JSON encoding for indexing and manual privacy review before any external synchronization or publishing actions (Workflow step 5).\n- [SAFE]: The skill follows secure coding practices, such as using
yaml.safe_load()in its validation tools and implementing a case-insensitive blacklist (e.g., secrets, tokens, API keys) in its profile manager to prevent sensitive data exposure. It also enforces restricted file permissions (0o600) on its local configuration storage.
Audit Metadata