The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including git status, git fetch, git restore, find, ripgrep, sed, sort, and tail. these tools are used to manage a specific local directory and extract meeting metadata from markdown files.
[EXTERNAL_DOWNLOADS]: The skill executes git fetch origin to synchronize the local meeting notes folder with a remote repository. This is an expected behavior for its stated purpose of maintaining a mirrored directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes meeting notes and transcripts from an external GitHub repository, which could contain malicious instructions designed to influence the agent's behavior during summarization or retrieval. * Ingestion points: Markdown files in the docs/meetings/granola/ directory (SKILL.md). * Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard embedded commands in the meeting notes. * Capability inventory: Shell command execution and Git operations (SKILL.md). * Sanitization: No sanitization or validation of the content of the meeting notes is performed before extraction.
[METADATA_POISONING]: The author metadata in the YAML frontmatter is set to openai, which conflicts with the identified author context itsAR-VR. This appears to be a template artifact rather than a malicious attempt at impersonation.

granola-github-meetings-sync