The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project, MCP, and plugin data retrieved from the Speakeasy platform, which could potentially contain malicious instructions.\n
Ingestion points: The workflow uses list_projects, find_mcp, and get_plugin to ingest data from the external platform environment in SKILL.md.\n
Boundary markers: The instructions explicitly mandate ending execution at confirmation points and asking the user before any write operations.\n
Capability inventory: The skill has the ability to register MCPs, create plugins, and modify platform distributions via tools like register_catalog_mcp and distribute_mcp_to_plugin.\n
Sanitization: The skill relies on mandatory user review and confirmation rather than automated sanitization.\n- [DYNAMIC_EXECUTION]: The workflow involves dynamic tool resolution and execution using call_platform_tool or an execute wrapper with computed tool paths.\n
Evidence: Instructions in SKILL.md specify running tools via call_platform_tool and using execute for multi-step reads on platform--<tool name> paths.\n
Mitigation: The skill requires the agent to stop for user confirmation before any write actions during dynamic execution.

migrate-mcp-between-projects