The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and follow terminology from local repository files (CONTEXT.md and CONTEXT-MAP.md).
Ingestion points: CONTEXT.md and CONTEXT-MAP.md are referenced as sources for ubiquitous language.
Boundary markers: Absent; there are no instructions to ignore embedded prompts within the context files.
Capability inventory: None; model invocation is disabled, and no tools for network, file writing, or command execution are requested.
Sanitization: Absent; the skill does not validate or sanitize the contents of the context files.

wait-what