The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user and agent to fetch content from an unverified third-party repository: https://github.com/mergisi/awesome-openclaw-agents.git. This source is not recognized as a trusted organization or well-known service.
[REMOTE_CODE_EXECUTION]: The documentation provides a workflow to git clone the external repository, perform npm install to download dependencies, and then execute the code using node bot.js. This creates a pattern where code from an unknown source is downloaded and executed locally.
[INDIRECT_PROMPT_INJECTION]: The skill architecture ingests SOUL.md files as untrusted data and interpolates them directly into the LLM system prompt.
Ingestion points: External configuration files (SOUL.md) are read via fs.readFileSync in SKILL.md (code examples).
Boundary markers: Absent; the content is passed directly as the system role content.
Capability inventory: The agents described have access to OpenAI API calls, file system writes (fs.writeFileSync), and subprocess spawning.
Sanitization: None; the content is used verbatim, allowing embedded instructions in a SOUL.md file to potentially override the agent's behavior.
[DYNAMIC_EXECUTION]: The skill includes code snippets for a 'multi-agent setup' that uses child_process.spawn to dynamically launch new Node.js processes based on file paths defined in an array.

awesome-openclaw-agents-templates