The Agent Skills Directory

[SAFE]: The skill provides administrative functionality for managing Speakeasy infrastructure with significant built-in guardrails. It mandates a multi-step verification process and requires human-in-the-loop confirmation for every write operation, ensuring that the agent does not perform unauthorized configuration changes. The instructions also contain a proactive safety directive forbidding the agent from requesting or transmitting sensitive credentials like API keys or OAuth tokens, following security best practices.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external API responses which represents a potential attack surface, though it is well-mitigated by the workflow design.
Ingestion points: Data is retrieved via tools such as list_projects, list_plugins, and get_mcp_connection_settings as described in SKILL.md.
Boundary markers: The agent is required to present a preview of all changes and obtain explicit user confirmation before execution.
Capability inventory: The skill possesses write capabilities through tools like set_mcp_address and set_mcp_network_access mentioned in SKILL.md.
Sanitization: The workflow incorporates freshness checks using expected_version and idempotency keys to ensure that modifications are consistent with the current state and user intent.

configure-private-mcp-access