The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a framework for ingesting data from external agents, which naturally creates an attack surface for indirect prompt injection where malicious instructions could be embedded in bot-provided content.
Ingestion points: Data enters the Paperclip context via external bot tool calls (e.g., paperclip_dot_tool) and signed events (e.g., mailbox_updated), which are then processed by the Runner (SKILL.md, runner-architecture.md).
Boundary markers: The skill provides specific instruction templates in onboarding-and-prompts.md designed to constrain bot behavior and clearly separate Paperclip-authorized operations from the bot's native capabilities.
Capability inventory: Implementation involves executing shell commands for testing and history (pnpm, git), network health checks (curl), and file system operations for task documentation (SKILL.md, conformance-tests.md, troubleshooting-and-deployment.md).
Sanitization: Security guidelines in runner-architecture.md mandate that filenames and contents from external bots be treated as untrusted data, and the avatar management section (onboarding-and-prompts.md) specifies re-encoding of images to strip metadata.
[COMMAND_EXECUTION]: The skill provides instructions for executing local shell-based tools for development, testing, and deployment.
Evidence: Mentions running pnpm exec vitest for conformance testing and executing scripts/deploy-stack.mjs to orchestrate Cloud deployments (conformance-tests.md, troubleshooting-and-deployment.md). These represent standard operations within the vendor's development environment.

create-external-bot-integration