agent-workflow-typesafe-ai
Standalone adapter turning bounded Agent-Workflow evidence into Jev Choice, Noul and Score receipts while the host keeps policy.
Overview
agent-workflow-typesafe is a standalone TypeSafe AI/Jev evidence adapter originally built for Agent-Workflow's external plugin boundary. It projects bounded, redacted task or skill evidence into typed questions, calls Jev, and normalizes the answers into Choice, Noul and Score receipts with explicit uncertainty and fallbacks.
The key boundary is that the host keeps authority: Agent-Workflow or another host owns thresholds, routing, execution, review and acceptance, while the adapter only returns advisory evidence. Version 0.1.2 declares compatibility with Agent-Workflow 0.10.0 to 0.10.2; newer Agent-Workflow releases include their own built-in TypeSafe provider, so this repository now serves mainly as a compatibility and evaluation surface.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which skill or route best fits this bounded task?(inferred) | — |
| Yes / no | Does the evidence support this claim?(inferred) | yes · no |
How it works
- Host passes redacted task or skill evidence
- Adapter builds typed questions and calls Jev (optional)
- Answers normalized into receipts with uncertainty
- Host applies its own thresholds and policy
Strengths
- Model never gets workflow authority
- Explicit uncertainty and fallbacks
- Published compatibility range
Caveats
- Only compatible with older Agent-Workflow versions
- Superseded by the built-in provider in 0.11+
At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Agent-Workflow plugin / library
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Standalone TypeSafe AI/Jev semantic-evidence adapter for bounded Agent-Workflow decisions, explicit uncertainty, compatibility evaluation, and host-owned policy.
hellogumbo/awesome-jev
Frequently asked questions
Does the adapter make decisions?
No, it returns advisory typed evidence; the host owns thresholds and actions.
Which Agent-Workflow versions does it support?
0.10.0 to 0.10.2; 0.11+ has a built-in TypeSafe provider.
Why keep it?
As a standalone evaluation surface and for hosts matching its compatibility contract.
Related projects
Codebase maintenance framework that turns issues into reviewed, tested changes via flows, listed with a Jev session checker.
ProjectbetaTypeScriptSoftware development framework for coding agents that hands dependency vetting, completion gates and debugging decisions to Jev.
LibraryAx, a DSPy-style multi-language LLM framework, listed with a native TypeSafe provider for Boolean, Choice, Score and raw Jev questions.
PluginproductionTypeScriptCline's official curated plugin collection, which includes jev-browser, jev-guard and jev-playground plugins.
PluginproductionTypeScriptComposio agent SDK with an optional TypeSafe provider that uses Jev to shortlist and pick tools and closed-set arguments.
ProjectproductionTypeScriptEffect TypeScript library whose @effect/ai-typesafe package maps classify, probability and rating operations to Jev questions.
ProjectproductionTypeScript