Skip to content

Code UX 0.9.8

Choose a tag to compare

View all tags
@numnx numnx released this 14 Jul 13:10
v0.9.8
436eee7
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Code UX 0.9.8

Code UX 0.9.8 advances the local-first agentic coding runtime with production Nodes automation, secure credential management, safe sprint rollbacks, faster and more restart-resilient orchestration, redesigned Tasks and Cost analytics, and broad dashboard, Chat, QA, and CI reliability improvements. See the release PR #3125 for the promotion record and complete validation evidence.

Release Snapshot

Item Details
Version 0.9.8
Release commit 436eee7173415281ef72d399533395f96de95bf0
Release PR #3125
Included dev sync #2968
Scope 776 files changed across runtime orchestration, Nodes and credentials, dashboard UX, documentation, tests, and repository hygiene

What’s New

Production Nodes Automation

Nodes now provides a durable, governed automation platform spanning authoring, execution, recovery, and inspection.

  • Unified agents, backend execution, validation, MCP, and the dashboard canvas on a canonical Graph v2 contract and node-definition registry.
  • Added durable, version-pinned publications and runs with resumable approvals, persisted attempts, bounded Foreach fan-out, redacted inspection, and authenticated headless recovery.
  • Added a TypeScript custom-node SDK with isolated build/runtime boundaries, immutable artifacts, declared capabilities, and governed HTTP access.
  • Connected server-backed, migration-safe flow authoring to agent attachments and MCP authoring tools.

Secure Automation Credentials

Automation secrets can now be managed and bound without exposing their values to graphs, prompts, APIs, logs, or run records.

  • Added write-only create, rename, validate, rotate, replace, restrict, promote, revoke, and binding workflows in Settings.
  • Added explicit secure local custody and fail-closed mounted-file, Vault, and KMS custody for remote/headless operation.
  • Added credential slots for node flows and custom dashboards, atomic versioned writes, compare-and-swap concurrency, revocation precedence, and separate management/use authorization.
  • Added bounded validation and explicit policy-denial and version-conflict responses across credential APIs.

Sprint Rollbacks and Planning Controls

Sprint delivery now has safer recovery paths and more predictable start behavior.

  • Added first-class rollback lineage, conservative safety assessment, isolated automatic reverts, and agent-assisted fallback for ambiguous or conflicting history.
  • Kept automatic remote rollbacks provider-free: they open a dedicated PR, wait for green checks, merge through the Git host, and complete only after merge confirmation; local Git projects use an equivalent guarded local path.
  • Added safe branch freshness preflight, an explicit Update Branch action, and planning before deferred, scheduled, or after-sprint dispatch.
  • Improved planning decomposition around repository ownership and risk, with ETA and bounded polling guidance in MCP responses.
  • Added durable Mark Completed and Mark QA Pass controls that stop active work and clear the matching handoff correctly.

Faster, Restart-Resilient Orchestration

High-concurrency work is faster while preserving durable ownership, publication, and recovery guarantees.

  • Added concurrent virtual-worker dispatch with durable leases, provider claims, repair ownership, heartbeats, and adaptive CPU/memory admission.
  • Reduced warm-launch overhead through transcript byte cursors, bounded buffers, tail-only persistence, reusable Docker inventory and immutable volumes, Git bundles, and runtime ownership reuse.
  • Added atomic publication retries so concurrent feature and repair writers preserve both histories.
  • Moved large backfills, index construction, pruning, and vacuum work out of startup into bounded persisted maintenance.
  • Preserved QA, CI-fix, and merge-conflict sessions—including model, workspace, baseline, checkpoints, and publication state—across forced runtime restarts.
  • Reconciled provider work interrupted during terminal bookkeeping without duplicate execution while keeping genuinely interrupted attempts recoverable.

Earlier Invocation Visibility and Reliable Chat

  • CLI invocations now appear from one persisted execution row during cancellable workspace preparation, then retain that identity through provider execution and finalization.
  • Stats → System consumes realtime execution updates without losing pagination, filters, or stable rows.
  • Chat continuations now use a durable reply-delivery barrier, suppress duplicate recursive wakeups, and inherit validated authenticated thread context.
  • Long speech begins after the first sentence-aware chunk, prefetches one chunk ahead, preserves playback order, and cleans up correctly after stop, mute, replay, or cancellation.

Tasks, Cost Analytics, and Delivery Status

  • Redesigned Tasks as a responsive, accessible, information-dense board with stronger filters, overview context, dependency rows, action menus, and executable browser coverage.
  • Added a dedicated Stats Cost mode with sprint/task spend, averages, token allocation, trends, model and execution-purpose breakdowns, and detailed ledgers.
  • Unified delivery status across Live, Tasks, Sprints, and Overview from Coding through PR, QA, CI, Merge, and Completion, with separate QA review evidence.
  • Refined sprint gallery cells so human attention, failure, intervention, and active work remain distinct and reduced-motion aware.

Dashboard Performance and Interaction Polish

  • Lazy-loaded noncritical shell and modal features, compacted Overview projections, removed duplicate reads, and reduced expensive refresh pressure.
  • In measured local production builds, cold DOM-ready improved from about 5.4 seconds to 159 ms, first content from about 5.7 seconds to 432 ms, Overview task payload from about 3.04 MB to 24 KB, initial JavaScript from about 1.39 MB to 567 KB, and initial CSS from about 1.15 MB to 659 KB.
  • Added sustained 3D Chat work-tool feedback and a transient response bubble with refined large-screen and reduced-motion behavior.
  • Expanded the accessible in-house select control, made new/imported project guidance defaults explicit, and refreshed the vendored models.dev catalogue.

Fixes

CI, QA, and Completion Reliability

  • Fixed historical cancelled or failed GitHub checks overriding newer successful reruns and redispatching completed coding tasks.
  • Fixed CI-owned and explicit handoff states collapsing back to pending coding or stale merge/guardrail attention.
  • Bounded sprint QA exhaustion, reserved the final cycle for verification, suppressed unverifiable task expansion, and deduplicated the resulting human handoff.
  • Fixed active runtime state overwriting manual completion and follow-up projections remaining stuck after successful work.

Nodes and Credential Reliability

  • Fixed legacy Nodes types and handles failing canonical validation, flattened definition details, full-page canvas rerenders, stale navigation results, and hidden tabs retaining WebGL/realtime/polling resources.
  • Fixed split credential metadata/envelope commits, overlapping rotation races, late revocation/restriction enforcement, and global allowlisted projects gaining unintended management authority.
  • Hardened mounted-key parsing, Electron key initialization, Vault/KMS key identity validation, broker wiring, request shapes, and lifecycle HTTP status mapping.
  • Fixed automatic local credential custody on Windows rejecting synthesized POSIX modes; trusted loopback CLI/dashboard runtimes now inherit the current user-profile ACL while retaining path, type, size, atomic-install, and race checks.

Security Hardening

  • Fixed a resource-exhaustion path by rejecting non-finite or out-of-range egress timeouts before network dispatch; accepted request timeouts are explicitly bounded to 1 through 120,000 milliseconds.
  • Fixed a subprocess path-injection path by canonicalizing working directories and requiring an existing real directory inside the user home, application directory, OS temporary directory, or an explicit CODE_UX_DIRECTORY_BROWSER_ROOTS entry before either inline or helper-process spawning.

Routing, Planning, and Agent Updates

  • Fixed base-agent updates failing to extract valid structured provider responses.
  • Fixed Route Mapping thinking overrides not reaching planning and reflection/retry calls.
  • Added Codex Sol-family max and ultra reasoning choices while preserving appropriate caps for other models.
  • Fixed sparse project or sprint route overrides discarding inherited provider fields and preserved Settings navigation across same-tab refreshes.

Chat, Speech, and 3D Activity

  • Fixed scheduled continuations firing before their source reply settled, recursively rescheduling, cancelling the source turn, or losing authenticated thread context.
  • Fixed long speech waiting too long to begin, stopping around the backend chunk limit, or leaking stale synthesis and object URLs after cancellation.
  • Fixed 3D Chat work tools ending before awaited replies and restored matching live messages, progress, and tool counts.

Rollback and Restart Recovery

  • Fixed automatic rollback worktrees becoming invalid across containerized Git remounts and settled rollback audit tasks becoming dispatchable provider work.
  • Fixed local Git projects being rejected by remote-oriented rollback preflight and deferred sprints starting from stale branches or before planning completed.
  • Fixed restart windows losing QA-to-fix handoffs, CI-fix checkpoints, merge-conflict sessions, provider-completion truth, or effective workspace identity.
  • Fixed resumed Docker snapshots accepting a volume whose Git checkout was interrupted; snapshot reuse now verifies HEAD and rebuilds invalid workspaces from the requested branch.

Dashboard Status, Loading, and Release Hygiene

  • Fixed Overview workflow evidence diverging from Tasks and Live, broad interventions appearing as human-owned attention, and delivery/QA overlays visually merging.
  • Fixed eager optional chunks, multi-megabyte initial read models, duplicate stylesheet output, and unnecessary execution refreshes on dashboard landing paths.
  • Updated cross-platform release E2E assertions for the redesigned Tasks heading and accessible lane names.
  • Removed seven tracked transient provider-message artifacts from the repository root.

Validation

Check Result
Typecheck & Lint Passed
Backend Tests & Coverage Passed; 5,791 tests, 7 skipped, all thresholds met
Dashboard Tests Passed; 2,295 tests, 1 skipped
Security Audit Passed; no known vulnerabilities
Playwright E2E Passed; all 18 Linux, macOS, and Windows project shards
CodeQL Analysis Passed; zero open exact-head alerts
Package Smoke Passed on Linux, macOS, and Windows
Orchestration Passed for Linux Docker and macOS/Windows Electron DAGs
Desktop Release Candidates Passed on Linux, macOS, and Windows

Additional local validation:

pnpm run ci
pnpm run check:docs-web
pnpm run test:orchestration:ci-dag
pnpm exec playwright test tests/e2e/navigation/dashboard-workflows.spec.ts --project=navigation
CODE_UX_SKIP_RELEASE_INSTALL_BUILD=1 node scripts/verify-release-install.mjs
git diff --check

Local CI passed with 5,791 backend tests (7 skipped), 2,295 dashboard tests (1 skipped), all coverage gates, a production build, documentation parity, a two-restart Linux Docker QA DAG, 15 UI-migration Playwright tests plus the three implicated credential/navigation regression specs, and an isolated packed installation of @codeuxai/codeux@0.9.8.

Included Change Set

Known Follow-Up

  • Nodes Workflow and Dashboard chat quick actions remain opt-in behind dedicated feature flags while rollout validation continues; the underlying Nodes and custom-dashboard capabilities remain available through their normal surfaces.
Assets 8
Loading