The Agent Skills Directory

[SAFE]: The skill is a collection of Markdown documentation, templates, and specifications (Open Knowledge Format). It contains no scripts (.sh, .py, .js), binaries, or tool definitions.
[INDIRECT_PROMPT_INJECTION]: The skill defines the framework for ingesting external data into knowledge bundles. It includes a specific "Trust Model" (found in references/trust-model.md) which explicitly identifies external data as a potential prompt-injection surface and instructs agents to treat bundle contents as data rather than instructions.
Ingestion points: Data ingestion is defined in references/trust-model.md and references/glossary.md as the core purpose of the OKF format.
Boundary markers: references/trust-model.md Section 6 provides a explicit instruction to treat content as untrusted data.
Capability inventory: No executable capabilities or tool access are requested or defined in this skill.
Sanitization: The skill documents the principle of "append-only-on-meaning" and provenance tracking to mitigate risks of silent data corruption or malicious overrides.
[COMMAND_EXECUTION]: No shell commands, dynamic context injection (!command), or subprocess calls were detected.
[EXTERNAL_DOWNLOADS]: No network requests, remote script fetches, or package installations are present.

kb