The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script for a daily version check.
Evidence: The script uses find to check for a local timestamp file (~/.claude/.foundry-version-checked) and curl to fetch the current version from https://foundry.thehorizonfoundry.com/api/version.
Purpose: This is a documented rate-limited check to notify the user if a newer version of the suite is available.
[EXTERNAL_DOWNLOADS]: The skill fetches a brand template and documentation from GitHub if local copies are unavailable.
Evidence: URLs target https://raw.githubusercontent.com/horizon-foundry/foundry/.
Purpose: Retrieval of static documentation and templates from the official project repository.
[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and applying instructions from a local BRAND.md file, which creates a potential surface for indirect injection if that file is compromised.
Ingestion points: BRAND.md is read to determine the brand voice mode and content.
Boundary markers: No specific delimiters are used to wrap the content of BRAND.md when it is processed.
Capability inventory: The skill uses the ingested rules to write and review user-facing copy; no critical capabilities (like remote execution) are triggered by the brand file content.
Sanitization: None. Content from the brand file is treated as authoritative for stylistic decisions.

brand-voice