The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses bunx to download and execute the algal-skills package from the npm registry at runtime.
[REMOTE_CODE_EXECUTION]: Executing the algal-skills package via bunx constitutes remote code execution, as the package's code is fetched and run on the host system during skill execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through its data ingestion capabilities.
Ingestion points: The web-fetch, repo-survey, git-digest, and search-slice programs ingest data from external URLs and local git repositories (file: SKILL.md).
Boundary markers: The skill documentation states that the agent receives a "compact typed report" and that "raw logs never enter context," which provides some structural separation but does not prevent the agent from obeying instructions embedded within that report.
Capability inventory: The skill utilizes the exec tool to run the ALGAL runtime, which in turn can execute git, gh, fetch, rg, and arbitrary subprocesses based on program manifests (file: SKILL.md).
Sanitization: There is no mention of sanitization, escaping, or filtering of the content ingested from the web or repository files before it is processed by the model.

algal