The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically email and LinkedIn replies via instantly__list_emails and grinfi__list_linkedin_messages. This creates an indirect prompt injection surface where external content could influence the agent's summary or recommendations.
Ingestion points: instantly__list_emails, grinfi__list_linkedin_messages (SKILL.md, outreach-review).
Boundary markers: The skill explicitly requires operator approval for decisions and reply sends ("decide gated", "Reply sends are gated").
Capability inventory: Access to paperclipai CLI, MCP tools for Paperclip and CRM writes, and email/message sending tools.
Sanitization: Not explicitly described, relying on human-in-the-loop validation as the primary defense.
[COMMAND_EXECUTION]: Documentation within the skill instructs the agent to install and execute system-level tools and MCP servers to facilitate communication with the Paperclip control plane.
references/docs/paperclip-operator/cli-contract.md includes commands for npm install -g paperclipai and npx -y @bbengamin/paperclip-mcp-server.
Configuration instructions for Claude Code include claude mcp add paperclip -s project -- npx -y @bbengamin/paperclip-mcp-server.
[EXTERNAL_DOWNLOADS]: The skill references and installs packages from the npm registry to establish its operating environment.
The skill downloads @bbengamin/paperclip-mcp-server and paperclipai from the npm registry. These are vendor-owned resources from the author.

outreach-review