The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local project using uv run at the hardcoded path /Users/mtm/pdev/taylormonacelli/benevolentbadger. This involves running local code that is not part of a standard package registry.
[DATA_EXPOSURE]: Hardcoded absolute paths (e.g., /Users/mtm/Documents/Obsidian Vault/ and /Users/mtm/pdev/...) expose the local username and directory structure of the host machine.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted YouTube transcripts and performs significant processing (cleaning, sectioning) before writing to the local file system and committing to a Git repository.
Ingestion points: YouTube transcript fetched via the benevolentbadger tool (SKILL.md).
Boundary markers: None present; the transcript content is directly cleaned and processed by the LLM.
Capability inventory: File system write operations, command execution via uv, and git commit operations.
Sanitization: The skill performs spelling and grammar correction but lacks sanitization to prevent the LLM from following instructions embedded within the transcript text.
[COMMAND_EXECUTION]: Automated git add and git commit commands are executed based on the processed transcript content.

youtube-transcript-note