The Agent Skills Directory

Command Execution: The skill executes system commands via shell scripts to probe the operational status of the sandbox. It interacts with tools like git, ssh-add, gh, and container runtimes (docker, podman) to verify that they are not functionally blocked. Executing these commands is necessary for the skill's purpose of identifying sandbox-related workflow failures.
Dynamic Execution: Several probe scripts utilize embedded Python code and dynamic loading to perform diagnostic tasks. probe-2-localhost-bind.sh and probe-3-container-gateway.sh execute Python snippets to test loopback connectivity and parse JSON status reports. Runtime execution of scripts is a security consideration, but here it is used to perform specific, locally-defined diagnostic logic.
Sensitive Path Access: The skill identifies and checks the readability of various configuration files and directories. It probes paths such as ~/.ssh/ for signing keys, SSH agent sockets, and agent-specific settings files like .claude/settings.json. Although accessing these paths is a potential security concern, the probes are read-only and targeted at resolving environment permission issues.
Indirect Prompt Injection Surface: The skill ingests data from local command outputs and configuration settings. Information retrieved from git config, gh api, and local .json settings is used to generate reports for the agent. This creates a surface where external content could influence the agent, though the risk is minimized by the skill's specific diagnostic scope.

isolated-setup-doctor