The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from external recipe websites and YouTube metadata (descriptions, titles, and thumbnails) to populate the recipe notes. It utilizes WebFetch and Playwright to retrieve this data.
[COMMAND_EXECUTION]: Executes a specific curl command to verify the availability of high-resolution YouTube thumbnails (hq720.jpg) before including them in the note frontmatter. This command is restricted to returning HTTP status codes and does not execute remote content.
[DATA_EXFILTRATION]: The skill reads local files and fetches from URLs. While it performs network operations, these are primarily for data ingestion from user-specified recipe sources. There is no evidence of sending sensitive local data (like environment variables or credentials) to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external URLs and YouTube descriptions, which is then processed by the agent. This represents an attack surface where hidden instructions in a recipe could attempt to influence the formatting process.
Ingestion points: External recipe URLs, YouTube video descriptions, and local markdown files.
Boundary markers: None explicitly used when parsing the external text content.
Capability inventory: File system read/write, network access (WebFetch, Playwright), and shell command execution (curl).
Sanitization: The skill applies strict formatting rules (lowercase start, removing periods, etc.) to the input, which provides some structural normalization, but it does not include specific sanitization for malicious prompts.

recipe-cleanup