The Agent Skills Directory

[SAFE]: The skill provides explicit instructions to prevent data exposure by requiring that API keys never be printed, persisted, or committed. It further mandates the redaction of authorization headers, prompts, and customer identifiers from all outputs.
[SAFE]: External references and dependencies, including the official Cohere TypeScript and Python SDKs, are sourced from official repositories and documented neutrally as intended functional resources.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent attack surface as it processes external repository code while having the capability to modify files.
Ingestion points: Reads repository code and configuration via Read, Glob, and Grep as defined in SKILL.md.
Boundary markers: Establishes strict "Approval Boundaries" in SKILL.md to prevent unauthorized model changes or side effects.
Capability inventory: Includes Write and Edit for file modification and WebFetch for documentation retrieval across all scripts.
Sanitization: Mandatory "Output" section in SKILL.md requiring redaction of keys, documents, embeddings, and environment data.

cohere-sdk-patterns