The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and summarize untrusted data from external URLs (including YouTube transcripts) and local files. This creates a risk of indirect prompt injection where an attacker could place hidden instructions inside a web page or file to manipulate the agent's behavior when the content is processed.
Ingestion points: Processes arbitrary URLs and local file paths (e.g., PDF) provided by the user or found in the environment.
Boundary markers: The instructions do not specify any delimiters or system prompts to warn the agent about ignoring instructions embedded in the ingested content.
Capability inventory: The skill executes the summarize CLI, which has the ability to read from the local file system and make network requests.
Sanitization: No sanitization or filtering of the ingested content is mentioned before it is passed to the language model for summarization.
[EXTERNAL_DOWNLOADS]: The skill metadata specifies the installation of a binary from a third-party Homebrew tap (steipete/tap/summarize). This introduces a dependency on an external repository that is not part of the platform's trusted vendor list.
[COMMAND_EXECUTION]: The skill relies on the execution of the summarize command-line tool. This tool performs network operations to fetch web content and accesses the local file system to read documents, representing a broad capability set that should be monitored.

summarize