Skip to content

Jev Workflows 0.4.0

Latest
Latest

Choose a tag to compare

View all tags
@integrate-your-mind integrate-your-mind released this 26 Sep 07:04
v0.4.0
b167e24
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Jev Workflows 0.4.0 adds typed batch decisions, persistent task context, and receipt-linked outcome records to the open-source Codex plugin.

  • Eight MCP tools cover tool, model, task, skill, context, strategy, result, and custom decisions. Independent Choice, Noul, and Score questions can share one bounded evidence state.
  • Scoped task context retains objectives, acceptance criteria, corrections, evidence references, and available candidate catalogs. Continue, replace, and reset have explicit semantics.
  • Ambiguous choices can abstain, including ranking mode. Single available candidates are evaluated against an explicit non-action option.
  • Completion checks evaluate individual acceptance criteria. Outcome records preserve caller provenance and derive provider IDs from existing receipts.
  • Accounting separates attempts, validated responses, and provider receipts. HTTP 403 no longer implies a bad credential; recognized Codex network-policy denials have a distinct diagnostic.
  • Twelve host lifecycle adapters and three standalone skills are included. There are no plugin-imposed daily, byte, or session quotas by default.

Validation: 152 automated tests on Node 22; typecheck, manifest checks, and generated-package parity; fresh native automatic-hook requests with validated HTTP 200/provider IDs; and independent Codex QA covering the repaired behaviors. A fresh 16-case live benchmark returned 14 correct dispositions, six abstentions, and no false-confident wrong decision. This is a small advisory evaluation, not a universal accuracy guarantee.

Install:

codex plugin marketplace add integrate-your-mind/jev-codex-plugin
codex plugin add jev-workflows@jev-workflows

Requires Node.js 22+ and your own TypeSafe API key. Automation is opt-in on fresh installs. Jev is advisory and does not grant permission, switch models, execute decisions, or independently prove completion.

The release is publicly installable through the Git marketplace and attached archives. Official OpenAI catalog acceptance and universal Plugins Directory approval are separate; this release does not claim either listing is live.

Hosted GitHub Actions could not start because GitHub reported an account billing lock. The local and native checks above passed; hosted CI is not claimed as passed. See VERIFICATION.md for evidence and limitations.

Documentation clarification (2026-09-26): Ranking mode returns disposition: "ranking" without applying conservative confidence/probability floors. For a selected Choice candidate, the compatibility recommendation field still contains its ID; it does not imply those floors passed. Explicit insufficient_evidence always abstains and omits recommendation. Current source and skill documentation clarify this existing behavior; release runtime and asset hashes are unchanged.

Assets 10
Loading