The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation includes commands for piped installation from the web (e.g., curl | bash and iwr | iex). While these target the tool's own domain (openclaw.ai), they represent a pattern of downloading and executing remote code without verification.
Evidence: curl -fsSL https://openclaw.ai/install.sh | bash in SKILL.md and references/openclaw-overview.md.
Evidence: iwr -useb https://openclaw.ai/install.ps1 | iex in SKILL.md and references/openclaw-overview.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to design and build workflows and skills, which involves ingesting user requirements and potentially external documentation. This creates a surface for indirect prompt injection if the ingested data contains malicious instructions.
Ingestion points: User requirements, external documentation, and existing SKILL.md files processed in SKILL.md and references/workflow-engineering.md.
Boundary markers: None identified in the prompt templates.
Capability inventory: The skill utilizes CLI tools (openclaw, clawdhub) and file system access as described in SKILL.md.
Sanitization: The skill includes explicit instructions for the AI to not bypass pairing, sandboxing, or audits, and to avoid storing secrets in prompts.
[COMMAND_EXECUTION]: The skill documentation describes and encourages the use of various CLI commands for managing agents and skills.
Evidence: openclaw onboard --install-daemon, clawdhub install spotify-controller, and openclaw security audit --deep in SKILL.md and references/openclaw-overview.md.

openclaw