Let Agent Verify Before Acting
Jay299792458 · reddit · 2026-07-14
This long post proposes a method to add a "brake" to AI agents: instead of letting the model infer whether information is sufficient, use a declarative Checklist to verify whether required information already exists.
The core idea is to shift execution decisions from model reasoning to structured state checks:
- Present → Continue
- Missing → Ask user or pause
The author believes this prevents the model from guessing when information is missing or repeatedly asking for known info. The judgment is written to state, and execution logic only reads the state without re-deriving.
They also emphasize this is crucial for MCP / Agent and Tool Provider division of labor:
- Provider knows what the tool needs
- Agent only knows current dialogue context
- Checklist acts as a contract between them
Finally, they summarize the approach as: replacing "knowledge-based inference" with "presence-based verification," treating the final JSON state as an auditable log recording what info is known, what's missing, who filled it, and why execution was allowed or blocked.
More from coding & agent
- Soft Clamp cuts tool-call overuse in multi-teacher distillation, from 13.7% to 9.0% — antgroup · 2026-07-21
- Agent harness memory loss and compaction are still a major usability problem — adityaag · 2026-07-21
- SpecJudge runs locally on Ollama to pick the right-sized AI model for your project — jokiruiz · 2026-07-21
- A developer maps out six design rules for CLIs that humans and AI agents can both use — yujiezha · 2026-07-21
- A coding-agent skill that forces ADHD-friendly, answer-first output — ayghri · 2026-07-21
- A set of agent skills for CAD, robotics, and hardware design — earthtojake · 2026-07-21