The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the installation of the paperclipai global package and the use of @bbengamin/paperclip-mcp-server via npx. These are vendor-provided resources necessary for the skill's integration with the Paperclip ecosystem.
[COMMAND_EXECUTION]: The skill uses the paperclipai CLI and curl to interact with the Paperclip control plane and API. All such operations are protected by an operational boundary requiring the agent to present proposed changes and wait for operator approval before execution.
[CREDENTIALS_UNSAFE]: The skill mentions inspecting ~/.paperclip/auth.json to verify authentication context. However, it explicitly instructs the agent to never print bearer tokens or secrets discovered in the configuration, maintaining credential confidentiality.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through its ingestion of wiki content and operator summaries.
Ingestion points: External markdown content retrieved via paperclip-wiki-fetch and operator clarification summaries.
Boundary markers: Robust instructions requiring the agent to "Ask for approval before mutating Paperclip" and present all proposed mutations for review.
Capability inventory: Capabilities include CLI mutations, MCP tool execution, and direct REST API calls (SKILL.md, integration-matrix.md).
Sanitization: The instructions mandate a verification step where existing state is read and proposed diffs are presented before any write operations.

paperclip-record-strategy