The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a custom binary named islandiguana located at the hardcoded local path /Users/mtm/go/bin/islandiguana to interact with the filesystem and query the Obsidian vault.- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted recipe ingredient lines and vault metadata (filenames) to perform semantic matching and generate new Markdown files, creating an injection surface.
Ingestion points: Recipe ingredient lines provided in the prompt context and filenames retrieved from the local vault via the islandiguana tool.
Boundary markers: The prompt construction in tests/build_prompt.py uses plain text labels like 'Recipe ingredient line:' without robust escaping or XML-style delimiters to isolate untrusted data.
Capability inventory: The skill is capable of executing shell commands (islandiguana) and writing new note files to the local Obsidian vault.
Sanitization: There is no evidence of input validation or sanitization for the ingredient text or the resulting file names/content.- [EXTERNAL_DOWNLOADS]: The test instructions in tests/README.md direct the user to run pnpm dlx promptfoo, which triggers the download and execution of the promptfoo package from the NPM registry at runtime.- [CREDENTIALS_UNSAFE]: Documentation for running tests includes a command (security find-generic-password -s ANTHROPIC_API_KEY -w) that retrieves an API key from the macOS Keychain in plaintext to be used as an environment variable. While this is a common developer pattern, it exposes local credential management to the execution context.

resolve-recipe-ingredient-link