Three thin runtime-control adapters for LangChain agents: trust gate, budget gate, and post-run receipt

Talking Shop
1

I have been working on a small set of runtime-control adapters for LangChain agents running on LangGraph.

The goal is not to build a full platform, but to make three narrow control points easier to discuss, test, and compose: one before tool execution, one before model execution, and one after the run.

The three pieces are:

  • god-spear-mcp-gate: a minimal preflight trust gate for MCP-style tool requests

  • token-governor-langchain-middleware: a thin budget-window middleware example before model execution

  • aro-audit-langchain-receipt: a compact post-run receipt emitter

I kept these as small and composable adapters instead of one package because each control point has a different runtime role:

  • trust gate before tool execution

  • budget gate before model execution

  • compact receipt after execution

The shape I am exploring is roughly:

request
→ optional trust gate
→ optional budget gate
→ execution
→ compact receipt

I am not presenting these as official integrations or production-ready platform components. They are thin adapter examples meant to make runtime control points easier to reason about.

I would appreciate feedback on one specific question:

For LangChain users, would these be better positioned as:

  1. example adapters,

  2. middleware patterns,

  3. integration-side utilities,
    or something else entirely?

If this is interesting, I can share the repo names and demo layout in follow-up replies.