Verdict
Local model verifying CLI outcomes for AI agents. Exit code 0 ≠ proof
Visit website →About
Verdict is a tiny machine learning model that reads a CLI tool call and returns success, failure, blocked, or unknown. Runs locally on CPU with no Python dependencies. Only 197 KB in size, with 0.54 ms median inference time.
Designed as a semantic outcome layer for coding agents, CI runners, and tool-using systems. It looks beyond the process exit code and reads the command, stdout, and stderr. Agents need that distinction before they retry, stop, or claim completion.
Evaluated on 192 synthetic examples from 48 CLI families entirely absent from training. Achieved 93.2% accuracy vs 47.9% for the exit-code baseline. Training used 1,440 synthetic examples from 60 different families.
Verdict is an advisory classifier, not proof that a task completed. Keep deterministic checks and authoritative state verification as the source of truth. Apache-2.0 license, synthetic dataset CC0-1.0.
AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.
Why we picked it
Verdict addresses a hidden problem in AI agent automation: exit code 0 doesn't mean the command actually succeeded. It might have run zero tests, returned HTTP 500, or produced incomplete output. This tiny model provides a lightweight, fast semantic verification layer that helps agents make better decisions.
Related tools
- 1 Rystic Realistic local API simulators for testing AI agents Freemium
- 2 Superfik Open-source canvas to watch AI coding agents work — each on its own card Free
- 3 Momentic AI QA engineer that catches bugs before they ship Freemium
- 4 Stroq Finds leaked secrets in AI coding agent sessions Free
- 5 Backenly Autonomous backend your coding agent builds over MCP and keeps running Freemium
- 6 Overmind Train specialized models from your agent's production traces Freemium