The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes codebase content and has powerful shell capabilities, creating a risk where malicious instructions in project files could be executed by the agent. Ingestion points include codebase search and file read tools like codex_read_file and read_file. Boundary markers and sanitization are absent, as the system prompt in agents/openai.yaml does not instruct the agent to ignore or delimit instructions found within the files it processes. The capability inventory includes bash access, codex_write_file, and openclaw_execute.
[COMMAND_EXECUTION]: The agent is granted the bash tool in agents/openai.yaml, which allows for arbitrary shell command execution. This capability could be misused if the agent is influenced by untrusted data from the project files it is instructed to read.
[EXTERNAL_DOWNLOADS]: The skill references and requires the openclaw CLI tool from a third-party repository (github.com/anomalyco/openclaw) that is not associated with an established or trusted vendor.

yolo-auto