cyberneticlibrary
← library

Approve or block AI coding decisions

thumbgatepluginsetup L2★3,035
davepoon/buildwithclaude ↗
What it does

Block agent actions via human disapproval gates

Best for

Learning guardrails from real agent failures in production with NIST/SOC2 audit trail

Inputs
  • · Agent action (any LLM output)
  • · Human thumbs-down (👎) feedback
Outputs
  • · Captured lesson
  • · Blocking rule
  • · Pattern prevention
  • · Budget/compliance tracking
Requires
  • · ThumbGate MCP server
Preconditions

ThumbGate MCP deployed, human feedback loop active, agent integration wired

Failure modes

Overfitting to single example, false positives blocking legitimate actions, rule explosion, slow feedback loop

Trust signals
  • · 33 pre-action gates
  • · NIST/SOC2 compliance tags
  • · budget enforcement