The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to bypass the security sandbox to run commands in the host context. Specifically, it directs the agent to set sandbox_permissions: "require_escalated" to overcome keychain authentication errors, network blocks, or file system restrictions.
Evidence: Instructions to run gh repo, gh pr, git add, git commit, and git push outside the sandbox when restrictions are encountered.
[PROMPT_INJECTION]: The skill provides instructions for the agent to escalate permissions and use host execution mechanisms, which overrides the default secure sandboxing behavior intended by the platform.
Evidence: Use the harness's official host-execution mechanism. In Codex, set sandbox_permissions: "require_escalated".
[INDIRECT_PROMPT_INJECTION]: The skill relies on reading the output of external CLI tools (git status, gh auth status, gh ... view) to determine state and authentication, creating an ingestion point for untrusted data that lacks boundary markers or sanitization logic.
Ingestion points: Output from gh auth status, git status -sb, git remote -v, and gh view commands.
Boundary markers: None present.
Capability inventory: Ability to execute shell commands (git, gh) with escalated host-level permissions.
Sanitization: None present.

github-outside-sandbox