Skip to content

AI-agent resources

Red-Govern publishes version-matched resources that help AI agents distinguish supported, conditional, and unsupported Amazon Redshift governance workflows.

Available resources

  • The portable Skill at agent-skills/red-govern/SKILL.md.
  • The Claude project Skill mirror at .claude/skills/red-govern/.
  • Repository instructions for generic agents, Claude Code, Gemini CLI, and GitHub Copilot.
  • A deterministic, versioned Skill archive with legal and attribution files.
  • Contract evaluation fixtures covering all 21 problem entries and seven safety, privacy, scope, and version boundaries.
  • An optional local stdio MCP adapter exposing four offline-safe, structured configuration and privacy tools.
  • An optional OpenAI Agents SDK adapter exposing the same four offline-safe operations as strict local function tools without model execution.
  • A remote metadata API contract defining four GET-only public metadata operations for future Actions and HTTP clients; the planned server is not deployed and Custom GPT Actions remain disabled.

All capability claims derive from the canonical problem map. Platform adapters must remain thin and must not invent commands or redefine capability status.

Important limits

Publishing these files improves interoperability and discoverability. It does not automatically install Red-Govern in every agent, force model indexing, guarantee search ranking, or provide unrestricted command execution.

The OpenAI Agents adapter also does not execute a model on construction.

The local MCP adapter does not connect to Amazon Redshift, execute SQL, write files, open a network port, or provide a hosted service. The OpenAI Agents adapter also does not execute a model on construction. Red-Govern does not perform destructive remediation and does not prove that an object is safe to delete. Never provide credentials or unredacted production outputs to an AI agent.