reviewing-meetings
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests meeting summaries, transcripts, and metadata from external sources.
- Ingestion points: Meetings are retrieved from external providers as described in references/source-interpretation.md.
- Boundary markers: Instructions in references/source-interpretation.md explicitly mandate treating retrieved material as untrusted data and ignoring any embedded instructions meant to redirect the workflow.
- Capability inventory: The skill can perform file writes via the Obsidian CLI, interact with task management systems, and create calendar entries.
- Sanitization: The skill requires normalization of source-derived values and manual approval for all durable changes, minimizing the impact of potential injection.
- [COMMAND_EXECUTION]: The skill invokes the Obsidian CLI for document management.
- Evidence: Instructions in references/source-interpretation.md enforce strict path sanitization, including the removal of path separators and traversal components, to ensure file operations remain within the designated meeting folder.
- [SAFE]: The skill enforces a read-only discovery phase and requires explicit, unambiguous user approval for every proposed action. It also implements a re-read before writing strategy to verify the state of the target system before applying changes.
Audit Metadata