The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands to manage remote testing infrastructure. Examples include crabbox warmup, crabbox run, pnpm crabbox:run, and blacksmith testbox list across Linux, macOS, and Windows environments.
[EXTERNAL_DOWNLOADS]: The skill references downloading and installing tools from external sources, including brew install openclaw/tap/crabbox and fetching pull request data from GitHub using the --fresh-pr flag.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data sources, creating a potential surface for indirect prompt injection.
Ingestion points: The agent is directed to read remote command outputs, JSON summaries (containing fields like leaseId and commandPhases), remote logs, and VNC capture data in SKILL.md.
Boundary markers: No explicit boundary markers or "ignore instructions" delimiters are specified for the external data being processed.
Capability inventory: The skill has access to extensive shell execution capabilities locally and on remote targets, including sudo operations and network-based CLI tools mentioned in SKILL.md.
Sanitization: The skill includes instructions for manual redaction of secrets in logs, but no automated sanitization for the content processed by the LLM.
[DYNAMIC_EXECUTION]: The skill utilizes features for dynamic script execution on remote hosts through the --script and --script-stdin flags, which involve uploading and running local files or strings in a remote shell environment.

crabbox