The Agent Skills Directory

[SAFE]: The skill provides instructional guidelines for the development and review of agent skills, focusing on structure, metadata, and portability.
[SAFE]: Explicit security requirements are included, such as the prohibition of hardcoded secrets, private credentials, internal hostnames, and absolute machine paths.
[SAFE]: The conventions mandate the use of execution gates and safety checks for any destructive or high-cost operations, ensuring the agent stops and reports ambiguous or unsafe scopes.
[SAFE]: It promotes a least-privilege approach by recommending that only necessary reference files be loaded into context and that subagents only receive the workflow relevant to their specific tasks.

skill-conventions