The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of instructions from other skills into the agent's current context.
Ingestion points: The execute function in handler.ts retrieves instructions from external skill definitions via the skill-registry.
Boundary markers: The retrieved instructions are returned within a structured JSON object under the instructions key, which helps the agent distinguish them from other tool outputs.
Capability inventory: This skill handler is restricted to string manipulation and registry lookups; it does not have the capability to write files, execute subprocesses, or perform network operations.
Sanitization: The handler enforces a 6,000-character limit (DOC_CAP) on instructions but does not filter the content for potentially malicious prompt injection strings, as the source is assumed to be the system's own skill registry.

skill-loader