The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill contains a shell script for a version check that queries https://foundry.thehorizonfoundry.com/api/version. This is a legitimate vendor resource used to notify the user of updates. It includes a local rate-limiting mechanism using a stamp file at ~/.claude/.foundry-version-checked to ensure the check only occurs once every 24 hours.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository being audited, which presents a potential injection surface.
Ingestion points: Reads repository files, PRODUCT.md, and project documentation during Step 1 and Step 3.
Boundary markers: The skill explicitly requires that the agent ground findings in code and use file:line citations.
Capability inventory: The skill has the capability to search the repository, read files, and write findings reports to the output or project documentation directories.
Sanitization: The skill enforces a rule that findings must be quoted from the source material; any candidate finding that cannot quote a line must be reported as a question rather than a fact, reducing the risk of the agent obeying hidden instructions.

feature-design