Qualixar Jev Codex Workbench 1.1.3

Qualixar Jev Codex Workbench 1.1.3

Version 1.1.3 adds standing, workspace-scoped Jev operation to the existing Codex workbench. Codex keeps its selected coding model and native execution permissions. The original 20 workflows and grant-based path remain available for unenrolled workspaces.

What changed

• One-time workspace enrollment with expiry and shared daily call/payload limits. Edits do not revoke enrollment; a changed request gets a fresh decision.
• Compact MCP decisions with private local receipts and exact jev_recall recovery.
• Extractive output reduction. Automatic prompt preparation stays local, and automatic output reduction runs only on a selected local Laya-MLX sieve route. Errors, mandatory instructions, and uncertain material stay available.
• Workspace-level cache and coalescing for identical simultaneous requests, shared with child work.
• Browser bridge for Codex's existing authorized tab, with approved origins, observed controls, a bounded step cap, and independent final verification. No replacement browser is installed.
• Optional pinned Laya-MLX runtime and resident worker for selected local recipes. The first local request after an idle broker restart now waits a bounded time for worker warmup, without a manual terminal command.
• Updated README with four architecture/flow PNGs, corrected installation and upgrade guidance, security boundaries, and a scoped validation note.

Validation

The clean release checkout passed 301 Python tests, all supplied fixture contracts, 29 Node browser tests, and the 350-file manifest check. The installed plugin returned 11 MCP tools. On the tested Mac, the installed browser bridge reused the same authorized tab, and a stopped broker restarted and served a fresh local Laya-MLX probe without manual warmup. An isolated home without SLM configuration served offline tools; SLM remained enabled and healthy on the development host.

These checks establish functional behavior on the tested setup. They do not establish a general token, time, cost, or subscription-quota saving. A Codex CLI treatment run was blocked by that run's native approval policy, so no valid matched host savings percentage is reported. See the validation note.

Installation and upgrade

Start with the README. Existing 1.1.2 users should follow the source-aware upgrade guide so an old marketplace source does not reinstall 1.1.2. Review changed Qualixar hooks, restart Codex, and enroll each trusted Git workspace once. Subfolders of one Git repository share its enrollment; a different repository needs its own policy. Native Codex permissions remain in force.

v1.1.2 — Codex Desktop trust UX

This release corrects the Desktop setup path for Jev Policy Mode. Codex Desktop does not currently expose the /hooks review browser. The README, installer, start guide, and new trust guide now direct users to a one-time Codex CLI review of only the three Qualixar Jev hooks: UserPromptSubmit, PreToolUse, and PostToolUse.\n\nPolicy Mode remains on by default in assist mode after those hooks are trusted. The release also documents the separate enterprise-managed deployment boundary: only an organization administrator using managed Codex policy and device deployment can configure managed hooks; a public plugin cannot and must not auto-trust itself.\n\nVerified: 170 Python tests passed with one managed loopback skip; package integrity verified 288 files; plugin and skill validation passed. No live provider calls were used for this documentation and installer UX correction.

v1.1.1 — Jev Policy Mode for Codex

Qualixar Jev Control v1.1.1 ships Policy Mode ON by default in assist mode. A local classifier routes bounded decisions without sending or persisting prompt text. Opt-in enforce mode can hold governed Bash and file-edit tools until a request-bound live Jev receipt matches the exact workspace, revision, case, classification, and request ID.\n\nThe release adds jev_policy_status and jev_policy_check, seven total MCP tools, installer mode configuration, three reviewable Codex lifecycle hooks, and failure-closed enforcement.\n\nVerified before release: 170 Python tests passed with one managed loopback skip; all 60 offline fixture contracts passed; 287 package files verified; plugin and skill validation passed; all 20 workflows completed through TypeSafe direct with jev-1.13.0. One initial case-04 response was rejected for a weighted-score schema mismatch and both retries passed. A request-bound live review of the implementation returned REVIEW in three runs; Jev remained advisory and execution_authorized stayed false.\n\nPolicy Mode is built to save Codex tokens by making narrow routing and ranking decisions before unnecessary context expansion or tool exploration. No universal percentage or guaranteed quality improvement is claimed without a controlled Codex-side benchmark.