The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads project source code and dependencies from external repositories. This includes fetching the platform repository from GitHub and downloading the paperclipai package from the NPM registry.- [REMOTE_CODE_EXECUTION]: The onboarding process utilizes automated execution of remote code via npx paperclipai onboard --yes, which downloads and executes a package from the NPM registry without user confirmation.- [COMMAND_EXECUTION]: The skill provides numerous shell commands for environment setup and runtime operations, including pnpm install, pnpm dev, and pnpm db:migrate. The platform also explicitly supports a bash runtime for agents.- [INDIRECT_PROMPT_INJECTION]: The orchestration architecture creates an attack surface where malicious data could influence agent behavior. 1. Ingestion points: Untrusted data enters the agent context through task.title, task.description, and the instructions field returned by the /heartbeat API endpoint. 2. Boundary markers: The provided implementation patterns do not show the use of delimiters or warnings to ignore embedded instructions in task data. 3. Capability inventory: The skill documentation describes agents with access to a bash runtime, http bot capabilities, and the ability to post tool call outputs back to the platform. 4. Sanitization: There is no evidence of input validation, escaping, or filtering of the external task content before it is processed by the agent runtime.

paperclip-ai-orchestration