The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses highly sensitive files and directories to diagnose authentication issues.\n
Evidence: Uses ls -la ~/.ssh and sed -n '1,220p' ~/.ssh/config to inspect configuration.\n
Impact: These files contain host definitions and identity file paths. While used for diagnosis, exposing this information to the agent context is a data exposure risk.\n- [COMMAND_EXECUTION]: The skill relies on executing system commands to inspect and modify the local environment.\n
Evidence: Commands like gh auth status, ssh-add -l, ssh -T, and git remote set-url are used to manage authentication and repositories.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to the ingestion of local configuration data.\n
Ingestion points: Reads configuration from ~/.ssh/config and the output of git remote -v.\n
Boundary markers: Absent.\n
Capability inventory: The skill can modify local git configuration using git remote set-url and git config.\n
Sanitization: Absent. The skill processes the output of system commands directly.

multi-gh