The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the third-party NPM package @gongrzhe/server-gmail-autoauth-mcp@1.1.11 and zod-to-json-schema@3.22.5 into the agent's Docker environment.
Evidence: Phase 2 instructions modify the Dockerfile to include pnpm install -g for these packages.
[COMMAND_EXECUTION]: The skill requires several shell commands to be executed on the host system to set up directories, write configuration stubs, and restart services.
Evidence: Commands include mkdir -p ~/.gmail-mcp, cat > ~/.gmail-mcp/..., and systemctl --user restart nanoclaw.
[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to read and process external data from Gmail, which creates a surface for indirect prompt injection attacks where malicious emails could influence agent behavior.
Ingestion points: Gmail inbox via search_emails and read_email tools.
Boundary markers: None specified in the tool implementation instructions.
Capability inventory: High-privilege actions including send_email, delete_email, modify_email, and create_filter.
Sanitization: No explicit sanitization or filtering of email content is mentioned before it is processed by the agent.

add-gmail-tool