The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the GitHub CLI (gh) for repository management, code search, and API interactions. It implements a strict 'least privilege' model, requiring explicit user authorization for any command that mutates remote state or accesses sensitive configurations.
[EXTERNAL_DOWNLOADS]: The skill includes a procedure for optionally fetching a 'peer skill' (agent-task-relay) from the author's verified GitHub repository. This process is protected by mandatory user confirmation, immutable commit-hash pinning, and strict validation of the fetched instructions before they are applied.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of malicious instructions embedded in GitHub repository content or API responses. It implements an 'Instruction Authority' policy that requires the agent to delimit untrusted data and prevents tool output from overriding the skill's core security instructions.
Ingestion points: Data enters the agent's context through gh command output, including repository source code, PR descriptions, and GitHub API response bodies (specified in SKILL.md).
Boundary markers: The skill mandates that untrusted content must be delimited and left unchanged, specifically forbidding it from granting authority or expanding scope.
Capability inventory: The skill possesses capabilities to execute shell commands via gh, write to temporary files for task descriptions, and perform network operations via the GitHub API.
Sanitization: The skill relies on strict instruction hierarchy and explicit delimiters to prevent data-driven instruction overrides.
[CREDENTIALS_SAFE]: The skill explicitly forbids handling secrets, tokens, or private keys directly. It instructs the agent to use only existing machine-local authentication and to provide commands for the user to run locally when sensitive credential operations (like gh secret) are required.

simple-github-cli