The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute remote shell and PowerShell scripts directly from the internet using the pipe-to-shell pattern. This matches a confirmed detection by automated scanners. Evidence: curl -fsSL https://openclaw.ai/install.sh | bash and iwr -useb https://openclaw.ai/install.ps1 | iex in SKILL.md.
[COMMAND_EXECUTION]: The skill encourages the installation of global software and system-level configuration, including background daemon setup. Evidence: npm install -g openclaw@latest and openclaw onboard --install-daemon.
[DATA_EXFILTRATION]: The skill reads the ~/.openclaw/openclaw.json file, which is documented to store sensitive secrets. Evidence: cat ~/.openclaw/openclaw.json reads credentials such as OPENROUTER_API_KEY, Telegram botToken, and Discord tokens, as documented in references/config-reference.md.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and index files from docs.openclaw.ai to dynamically determine agent behavior and configuration steps.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by processing external untrusted content to guide its operations. 1. Ingestion points: Fetches llms.txt and specific documentation pages from https://docs.openclaw.ai/. 2. Boundary markers: None present to isolate external data from instructions. 3. Capability inventory: Shell execution (bash), file reading/writing (cat, config edits), and networking (curl). 4. Sanitization: No input validation or sanitization is performed on the fetched content.

openclaw-config