The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a standalone Node.js script (okf-validate.mjs) for checking bundle conformance. The script performs read-only operations on the local file system to parse markdown frontmatter and check link connectivity. It has zero external dependencies and does not perform network operations.
[INDIRECT_PROMPT_INJECTION]: As a tool for processing external knowledge bundles (markdown files), the skill handles an inherent attack surface for indirect prompt injection. However, it implements robust mitigation strategies:
Ingestion points: Reads markdown files and YAML frontmatter from user-provided directories (SKILL.md, okf-validate.mjs).
Boundary markers: Uses structured YAML frontmatter and conventional markdown headings to delimit data from instructions.
Capability inventory: Includes file read/write operations (staged for review) and local script execution for validation.
Sanitization: Employs a 'sources' attribution family for all factual claims, human-in-the-loop review cycles for writing, and 'Attested Computations' which strictly forbid the agent from authoring or editing executable code (e.g., SQL).
[SAFE]: The skill references official documentation and repositories from Google Cloud Platform (github.com/GoogleCloudPlatform/*), which is a well-known and trusted organization. These references are used for specification guidance and provenance tracking.

okf