agent-handoff-gate
Experimental protocol that checks the evidence behind a worker agent's PASS or BLOCKED report, with Jev answering narrow evidence questions.
This repository is mostly documentation or write-up rather than runnable code.
Overview
Agent Handoff Gate is a specification and offline evaluation kit for delegated coding work. When a worker agent reports PASS or BLOCKED, the gate compares the task's requirements with the candidate and its evidence before the result reaches the lead agent. It addresses two common failures: tests that pass without exercising the requested behaviour, and workers that stop with a usable fix because their repair budget ran out.
Exact checks stay in code. A semantic evaluator such as Jev answers narrow questions about what the evidence actually supports, and a separate policy decides whether to prepare a review packet, request an already authorized continuation, or return the issue to the lead. Test results, evaluator answers, authorizations and the lead's acceptance stay separate records. Version 0.1.0 includes schemas, worked examples, a TypeSafe/Jev integration contract and a benchmark, but no production controller, sandbox or runtime adapter.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does the evidence actually show the requested behaviour was tested?(inferred) | yes · no |
| Yes / no | Does this BLOCKED report contain a workable local fix worth one bounded continuation?(inferred) | yes · no |
How it works
- Worker submits PASS or BLOCKED with evidence
- Code runs exact checks against task and evidence schemas
- Jev answers narrow questions about what the evidence supports
- Policy picks: review packet, authorized continuation, or return to lead
- Lead performs the final review
Strengths
- Separates evidence, evaluation, authorization and acceptance
- Keeps exact checks deterministic
- Includes a reproducible offline benchmark
Caveats
- Experimental spec only; no controller or agent adapter yet
- Included tools do not call any API
Getting started
Start with docs/walkthrough.md and the protocol in docs/protocol.md.At a glance
- Type
- Project
- Status
- research
- Language
- Python
- Runtime
- specification + offline evaluation kit
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Experimental protocol for evidence-aware agent handoffs, with Jev-assisted review before results reach the lead agent.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev do here?
It answers narrow questions about whether a worker's evidence supports its report; policy and the lead decide what happens.
Can I plug it into my agent today?
Not directly; version 0.1.0 has no runtime adapter or controller.
What problem does it target?
Worker reports whose tests never exercised the requested behaviour, and blocked workers that already have a workable fix.
Related projects
Software-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonClaude Code Stop hook that blocks an unverified done: one four-question Jev call only when files changed and no check passed.
PluginbetaTypeScriptClaude Code plugin where Claude plans phased work with checks and, at each stop, Jev judges whether the work is really done.
PluginbetaTypeScriptStop hook for coding agents where Jev checks each attempted stop against plain-language rules and sends the agent back if one is violated.
HookbetaPythonAsynchronous System-1 supervisory plugin for Hermes Agent with typed Jev decisions, ranking, verification and an opt-in tool gate.
ProjectbetaPythonClaude Code Stop hook that uses Jev to catch turns ending half-done: no action, broken promises, unwatched work or wrong language.
HookbetaPython