soia-pkm-maintain
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill modifies AI agent configuration files, specifically .claude/settings.json and config.toml, to register hooks for session logging.
- [SAFE]: Instructions explicitly mandate that the agent must obtain user consent before making any configuration changes, preventing silent persistence.
- [DATA_EXFILTRATION]: The skill reads markdown files and metadata within the user's Obsidian vault for linting and organizational purposes.
- [PROMPT_INJECTION]: The skill ingests untrusted vault content via scripts/lint_vault.py (Ingestion point) which is processed by the AI for reporting without defined boundary markers (Sanitization absent), creating an indirect injection surface that interacts with local file-writing capabilities (Capability inventory).
- [SAFE]: No network exfiltration or malicious intent was detected; all operations are consistent with the skill's stated maintenance purpose and limited to the local file system.
Audit Metadata