The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses bunx to download the algal-skills package from a public registry during its execution.
[REMOTE_CODE_EXECUTION]: The skill executes the algal-skills package using bunx. This represents the execution of code that is not locally present or sourced from a recognized trusted provider.
[COMMAND_EXECUTION]: The skill performs shell command execution to run the bunx utility for its core functionality.
[INDIRECT_PROMPT_INJECTION]: The skill fetches data from arbitrary external URLs and injects the processed text into the agent's context, providing a vector for indirect prompt injection attacks.
Ingestion points: Content is fetched from external web pages via the URL provided in the skill arguments.
Boundary markers: The implementation lacks explicit delimiters or instructions to the agent to treat the fetched content as untrusted data.
Capability inventory: The skill environment permits the use of the exec tool, which could be misused if an injection attack succeeds.
Sanitization: The skill strips script and style tags from the HTML, which mitigates certain types of malicious scripts but does not prevent natural language prompt injection.

algal-fetch