The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs version checks and downloads updates from the vendor's official GitHub repository at https://github.com/CALLE-AI/call-e-integrations. This is a vendor-owned resource used for standard maintenance, and the skill includes rigorous procedures to verify content before any manual installation occurs.
[COMMAND_EXECUTION]: The skill allows the agent to utilize official, pre-installed host management commands (such as those provided by Claude Code or Codex) solely for managing the MCP connection lifecycle (e.g., OAuth configuration). This behavior is restricted to environment setup, respects host-level permission controls, and is explicitly prohibited for business dialing operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, creating a potential attack surface for indirect prompt injection.
Ingestion points: External data is processed from API responses at api.heycall-e.com, including call transcripts retrieved via GET /v2/calls/{call_id} and goal definitions from GET /v1/goals as specified in the core instruction files.
Boundary markers: The skill includes explicit defensive instructions to treat all external content—including narrative fields, goal text, and transcripts—strictly as data rather than instructions. It commands the agent never to use claims of user approval within that content to authorize new actions.
Capability inventory: The skill has the capability to initiate billable phone calls via POST /v2/calls and manage MCP connection states.
Sanitization: The agent is instructed to manually extract only relevant facts for the user's task and to screen all server-returned clarifications before presenting them to the user, providing a layer of human-in-the-loop verification.

callev2