The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted markdown files from a public-facing inbox directory.
Ingestion points: Reads raw .md files and metadata from /Users/ikatkov/Obsidian/Default/inbox/.
Boundary markers: No delimiters or 'ignore embedded instructions' warnings are present to isolate the processed note content from the agent's instructions.
Capability inventory: The skill can write and delete files, and execute shell commands based on classification logic.
Sanitization: There is no evidence of sanitization for the input data before it is used to determine classification or filename slugs.
[COMMAND_EXECUTION]: The skill executes shell commands to perform file lookups and image conversions.
Commands include ls, find, and magick (ImageMagick).
The logic interpolates generated slugs and filenames into commands, such as find /Users/ikatkov/Obsidian/Default -name "<new-filename>", which could be exploited if malicious filenames are introduced into the inbox.
[EXTERNAL_DOWNLOADS]: The skill fetches data and media from external sources to enrich notes.
Retrieves metadata from well-known services including IMDb (movies) and Goodreads (books).
Utilizes external tools like yt-dlp and instagram-gallery-download to fetch media content from social platforms for 'travel' and 'poses' categories.

obsidian-inbox