wiki-ingest
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-authored Obsidian notes (journal entries). This presents a surface for indirect prompt injection if a note contains instructions designed to manipulate the agent. However, the skill includes several robust mitigations:
- Ingestion points: The agent reads content from journal notes (
journal/YYYY-MM-DD) and other pages in the Obsidian vault. - Boundary markers: Processing is constrained to specific Markdown headings that follow a wiki-link pattern (e.g.,
# [[Name]]). - Capability inventory: The skill uses the Obsidian MCP server for file reads/writes and the
periodictool for listing files. It does not have general network access or arbitrary command execution capabilities. - Sanitization: The instructions require the agent to summarize research and extract only specific facts or intent rather than performing verbatim copying. Furthermore, a mandatory confirmation step ('Confirm' procedure) ensures the user reviews all proposed changes before they are applied.
Audit Metadata