The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill utilizes sudo to perform administrative system actions, such as starting and enabling services (sudo systemctl start docker) and installing system-level software (sudo apt-get install -y <package>). These operations are performed autonomously without requesting explicit user permission.
[CREDENTIALS_UNSAFE]: The instructions direct the agent to read the .env file to identify missing API keys. Although intended for configuration diagnostics, .env files are sensitive locations that commonly store credentials and secrets.
[COMMAND_EXECUTION]: The skill has broad capability to execute arbitrary shell commands via the exec tool to resolve errors. This includes terminating system processes (kill -9 <PID>), deleting authentication directories (rm -rf .beta/whatsapp-auth), and modifying file system permissions or database files.
[EXTERNAL_DOWNLOADS]: The skill performs dynamic installations of external software and dependencies from public registries using npm, npx playwright, apt-get, and Homebrew based on runtime error patterns.
[INDIRECT_PROMPT_INJECTION]: The skill's repair logic is triggered by parsing tool error messages, which creates a vulnerability surface.
Ingestion points: Tool error output (e.g., from Playwright or npm) which can be influenced by external data or malicious web content.
Boundary markers: None; the agent is instructed to act immediately on the error content.
Capability inventory: Full exec access with sudo privileges, along with read and write capabilities in SKILL.md.
Sanitization: There is no validation or sanitization of the package names or error strings before they are passed into shell commands like npm install <package-name> or apt-get install <package>.

self-heal