The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several shell commands including which, grep, mcporter, and openclaw during Phase 1 to inspect the local environment, check existing software versions, and query the configuration file ~/.openclaw/workspace/TOOLS.md.
[EXTERNAL_DOWNLOADS]: Phase 2 involves the use of established package managers such as Homebrew, npm, and pip to download and install external software. While these are well-known services, this capability allows the agent to pull and execute third-party code based on user input or external links.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in Phase 0 and Phase 1, where it ingests untrusted data from external tool documentation and local session history/logs. This data could contain malicious instructions designed to bypass the 'Phase 1' decision logic or trick the agent into recommending an unsafe tool.

installer-tester