The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to read the AGENTS.md file from the user-provided plugin checkout and execute the command found within it to bring up the environment (e.g., npm run wp-env start or docker-compose up -d). Because the content of AGENTS.md is controlled by the external plugin author and is executed without validation or a whitelist of allowed commands, this represents a significant command injection and RCE vector.
[COMMAND_EXECUTION]: The skill makes extensive use of powerful shell utilities including rg, grep, awk, and wp-cli. It specifically directs the agent to execute complex PHP logic via wp eval and wp eval-file, which provides full access to the underlying system and database when running in runtime mode.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection via the audited plugin's source code and the audit.yaml file.
Ingestion points: AGENTS.md (command strings) and audit document YAML blocks (seeding instructions).
Boundary markers: Absent; the instructions tell the agent to "follow it as documented" or use "representative input" based on the file content.
Capability inventory: Full shell execution (bash), PHP execution (wp eval), and filesystem access.
Sanitization: None provided; the skill relies on the agent to interpret and execute strings directly from the untrusted project files.

wp-abilities-verify