The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's central functionality involves executing shell commands and scripts on remote Linux, macOS, and Windows environments via the crabbox run and pnpm crabbox:run commands. This is used for CI-parity checks, broad testing, and E2E validation.
[INDIRECT_PROMPT_INJECTION]: The agent is instructed to process data from external GitHub Pull Requests using the --fresh-pr flag and to analyze outputs from remote test execution. This creates a surface where malicious instructions in external code or test results could influence the agent's behavior.
Ingestion points: GitHub PR file content, remote terminal outputs, and command logs (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are provided for parsing external PR content or remote logs.
Capability inventory: The agent can execute remote commands, write local files/logs, and modify environment configurations.
Sanitization: While the instructions emphasize redacting secrets using sed, there is no explicit instruction to sanitize natural language commands from processed external data.
[PRIVILEGE_ESCALATION]: The documentation includes instructions for using sudo to create symbolic links in system-wide directories (e.g., sudo ln -sf ... /usr/local/bin/) as part of a manual environment handoff repair pattern.
[DYNAMIC_EXECUTION]: The tool supports uploading and executing local scripts on remote targets via the --script and --script-stdin flags. This allows for runtime execution of dynamically generated or selected logic on the remote system.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the Crabbox CLI via a vendor-provided Homebrew tap (brew install openclaw/tap/crabbox) and refers to the coordinator service at https://crabbox.openclaw.ai.

crabbox