The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes meeting titles, generated notes, and transcripts which are categorized as untrusted data that could contain adversarial instructions.
Ingestion points: references/source-interpretation.md specifies the ingestion of meeting titles, summaries, transcripts, and participant text from external providers.
Boundary markers: While the instructions emphasize treating input as plain data, they do not define specific structural delimiters (e.g., XML tags or unique markers) to isolate external content within the model's context.
Capability inventory: The skill possesses capabilities for file modification (via Obsidian CLI), task/issue creation in canonical systems, calendar management, and drafting communications, as detailed in references/action-routing.md.
Sanitization: references/source-interpretation.md provides explicit sanitization logic, including whitespace normalization, control character removal, and strict path traversal checks to ensure filenames remain within a configured folder.
[COMMAND_EXECUTION]: The skill utilizes the Obsidian CLI for querying and writing meeting notes.
Evidence: SKILL.md and references/applying-approved-actions.md mandate the use of the Obsidian CLI for all note-related operations, including vault and target selection.
[EXTERNAL_DOWNLOADS]: The skill references and retrieves data from the meeting summarization service Granola.
Evidence: references/source-interpretation.md defines logic for mapping Granola native IDs to canonical URLs (https://notes.granola.ai/d/{native_id}) for data retrieval.

reviewing-meetings