AtomLane v0.15.0 — Visible Task Preflight & MCP Compatibility

AtomLane v0.15.0 fixes the repeated MCP App load failure reported as resources/read ... Unexpected response type and makes selective acceleration assessment visible in Codex Hooks.

What changed

• Historical versioned indicator resources, including ui://widget/atomlane-indicator-0.13.0.html, now resolve through the current server.
• Malformed params, unknown resources, and internal resource failures now use protocol-correct top-level JSON-RPC errors instead of tool-result payloads.
• A bundled, read-only UserPromptSubmit Hook now shows one of three advisory outcomes on every task: direct path, inspect at the execution boundary, or likely parallel candidate.
• The Hook never scans the project, executes commands, blocks the prompt, or declares concurrency safe. Python starts with -I -S; native Windows uses a tested .cmd launcher.
• The root Agent Plugins manifest was removed because current Codex releases classify it separately and suppress bundled lifecycle Hooks. The Codex-native package remains complete, and root mcp.json remains available for optional portable local-stdio use.

Upgrade

codex plugin marketplace upgrade atomlane
codex plugin add atomlane@atomlane

Start a new Codex task, open Hooks, review the AtomLane command, and trust it once. Existing tasks may retain old MCP App metadata and should be replaced by a new task after upgrading.

Source-bound verification

Source commit: fd8d5ca0e8a85d63b3eac30076786b984a52d7ae
Release commit: 47bb6d84c9a0fe871226b6773e2bd759f01fbff5 (docs-only descendant)

• Native Windows Preview: 240/240 tests passed; 9/9 release checks passed.
• macOS five-minute benchmark: 4 × 310-second tasks; 310.27s wall, 1240.33s serial equivalent, 930.06s saved this run, 3.9976× observed speedup; 10230.52s cumulative saved.
• Native Windows five-minute benchmark: 4 × 310-second tasks; 310.39s wall, 1240.00s serial equivalent, 929.61s saved this run, 3.9950× observed speedup; 8366.24s cumulative saved.

Serial equivalent is the sum of observed independent task runtimes; tasks were not rerun serially. These controlled low-load fixtures demonstrate overlap, not a universal speedup guarantee.

Live report: https://cloudguo123.github.io/atomlane/

Verify all attached assets with SHA256SUMS.

AtomLane v0.14.0 — MPL-2.0 Community Release

AtomLane v0.14.0 is the first release whose current project-authored source is distributed under MPL-2.0. The current community release stays open source and free for personal, research, educational, and commercial use, with no AtomLane account or payment required.

License boundary

• Current project-authored source is licensed under MPL-2.0 unless a file states otherwise.
• AtomLane v0.13.0 and repository history through a25827d retain their original MIT grants permanently.
• The historical MIT text is preserved in LICENSES/MIT-legacy.txt.
• Third-party components retain their respective licenses.
• Possible future alternative commercial licensing or separately licensed capabilities do not change the rights already granted for this release.

Governance clarity

• Added LICENSING.md, NOTICE, and TRADEMARKS.md.
• Contributions are accepted under MPL-2.0 without copyright assignment.
• No existing community capability was moved behind an account or paid tier.

Compatibility

• Plugin ID and install selector remain atomlane.
• macOS remains Stable; native Windows remains Preview.
• Atomic planning, live progress, per-run and cumulative savings, scenario routing, resource-aware scheduling, and the static Python parallelism advisor remain available.

Source-bound verification

Benchmark source commit: ca82bfa. Release commit: 56ca6f7, a docs-only descendant of that source.

• Native Windows Preview: 230/230 tests passed, including 20/20 native runtime checks and 9/9 release gates.
• macOS: four independent 310-second workloads; 310.168172 s parallel wall, 1240.135236 s serial equivalent, 929.967064 s saved this run, 3.9983x observed speedup, 9300.462878 s cumulative saved across 10 verified runs.
• Native Windows: four independent 310-second workloads; 310.411943 s parallel wall, 1240.003881 s serial equivalent, 929.591938 s saved this run, 3.9947x observed speedup, 7436.624030 s cumulative saved across 8 verified runs.

Serial equivalent is the sum of observed independent task runtimes; tasks were not rerun serially. These controlled low-load results demonstrate scheduler overlap for the stated fixtures, not a universal speedup guarantee.

Live report: https://cloudguo123.github.io/atomlane/

Verify every custom asset with SHA256SUMS.

AtomLane v0.13.0 — Final MIT release

AtomLane v0.13.0 is the final release of AtomLane distributed under the MIT License.

License boundary

• This release and repository history through commit a25827d remain available under MIT.
• Those grants are permanent and are not revoked by the license used for later versions.
• AtomLane v0.14.0 and later project-authored source use MPL-2.0 unless a file states otherwise.

Release state

This release preserves the universal macOS and native Windows Preview implementation, source-bound verification artifacts, live progress, per-run savings, cumulative savings, atomic planning, and the static Python parallelism advisor from the 0.13 line.

The attached benchmark evidence predates the final Windows statistics-lock race fix and remains bound to its recorded source commit; it is retained for auditability rather than presented as evidence for later code. Verify every download with SHA256SUMS.

AtomLane v0.12.0 — Native Windows Preview

AtomLane v0.12.0 introduces a fail-closed native Windows Preview while keeping macOS as the stable execution target.

Highlights

• Compiles supported work into a typed Atom IR with immutable plan hashes, control/data/lifecycle edges, artifact-conflict checks, and CPU/memory admission.
• Adds native Windows resource discovery, staged Job Object process-tree containment, UTF-8 MCP pipes, optional ConPTY output capture, PowerShell file atoms, and Windows path semantics.
• Keeps long runs visibly live with elapsed time, running/ready/completed/failed counters, per-run estimated savings, and cumulative savings. Captured task stdout/stderr is returned when each task completes.
• Adds a bounded, non-importing Python advisor for high-confidence ordered-map candidates. Rewrite previews are source-hash-bound review artifacts, not automatically verified speedups.
• Expands the release suite to 221/221 passing tests on native Windows, including 20/20 native runtime gates and 9/9 release checks.

Windows Preview boundary

The Job Object covers the waiting supervisor, target, and normally inherited Windows descendants. Target creation is staged after supervisor containment; it is not claimed as atomic target creation. WSL, Docker, WMI, services, scheduled tasks, and other brokered work remain separate execution realms. ConPTY is output-only in this Preview and merges stdout/stderr; explicit stdin, including an empty string, is rejected before target creation. The windows-2025 release runner is not proof of Windows 11 Desktop UI compatibility.

Verified five-minute evidence

Benchmark source commit: 61faa2b70a3495dab5bc0923b413d43643d788e7. Release commit: 7184a80f11fe13cbe3ce0335be7d9f3322f288f3, a docs-only descendant of that source.

• macOS: four independent 310-second workloads; 310.172480 s parallel wall, 1240.075514 s serial equivalent, 929.903034 s saved this run, 3.9980× observed speedup, 6510.533744 s cumulative saved across 7 verified runs.
• Native Windows: four independent 310-second workloads; 310.375690 s parallel wall, 1240.003885 s serial equivalent, 929.628195 s saved this run, 3.9952× observed speedup, 4648.032860 s cumulative saved across 5 verified runs.

Serial equivalent is the sum of observed independent task runtimes; the tasks were not rerun serially. These controlled low-load results demonstrate scheduler overlap, not a universal speedup guarantee for arbitrary projects.

Live report: https://cloudguo123.github.io/atomlane/

Install instructions and the complete support contract are in the README and docs/WINDOWS_PREVIEW.md. Verify downloads with SHA256SUMS.

AtomLane v0.11.0 — Safe Python Parallel Refactors

AtomLane now adds a bounded, non-executing Python parallelization advisor for long-running ordered-map workloads.

Highlights:

• Conservative dependency, effect, GIL/native, macOS spawn, resource, and benefit analysis
• Source-hash-bound review previews for the narrow proven-safe CPU subset
• Advisory routing for read I/O, network, subprocess, and native-owned work
• New optimize-python-parallelism skill and MCP tool
• Pre-parse token and line budgets that bound AST construction memory
• Public visual evidence: 113/113 tests and 7/7 release gates
• Retained five-minute benchmark evidence with live progress and measured savings

The advisor never imports, executes, or modifies analyzed target modules. Rewrite previews remain review-only and require runtime differential validation before performance claims.

AtomLane v0.10.1 — clean public rename

AtomLane now uses the canonical public repository slug cloudguo123/atomlane across source, Pages, installation commands, badges, support links, launch materials, citations, generated reports, and growth metrics. Existing installations remain compatible because the plugin ID, marketplace name, MCP key, event names, environment variables, and statistics directory remain mac-parallel-accelerator; GitHub redirects the previous repository URL.

AtomLane v0.10.0 — proven-safe parallelism

Mac Parallel Accelerator is now AtomLane.

Parallelize only what is proven safe.

• New AtomLane identity across Codex, Agent Plugins, MCP App, skill UI, documentation, Pages, and launch materials.
• Redesigned listing logo, social preview, and live-execution demo.
• Existing plugin ID, install commands, repository URL, MCP key, environment variables, and cumulative savings storage remain compatible.
• Verified by 44/44 tests, CI, CodeQL, HOL Plugin Scanner, and the retained controlled 5+ minute benchmark.

Install remains:

codex plugin marketplace add cloudguo123/mac-parallel-accelerator
codex plugin add mac-parallel-accelerator@mac-parallel-accelerator

Live verification report · Brand guide

v0.9.4 — portable discovery and first-run measurement

Mac Parallel Accelerator now ships as both a native Codex plugin and a schema-valid Agent Plugins 1.0.0 package.

Highlights:

• portable root plugin.json and local-stdio mcp.json, validated against the official 1.0.0 schemas
• privacy-reviewed first-run feedback form and aggregate conversion counters on the public report
• official listing packet, square artwork, terms, and additional no-cost discovery paths
• 43/43 behavioral regression tests, end-to-end self-test, reproducible UI bundle, Ruff, HOL scanner, and Pages deployment all passing
• retained controlled benchmark: 20m41s serial equivalent, 5m10s parallel wall, 15m31s saved, 4.00× observed

The benchmark serial equivalent is the sum of observed independent task runtimes, not a separately rerun serial workflow. Results are workload-specific.

v0.9.3 — marketplace security hardening

What changed

• Removed the bundled Zod dynamic-evaluation capability probe from the MCP App UI. The build and regression suite now reject both eval() and new Function() in the published bundle.
• Added a SHA-pinned HOL Plugin Scanner workflow with the public-marketplace profile, an 80-point minimum, and a high-severity failure gate.
• Added privacy metadata, .codexignore, manifest discovery links, explicit skill licensing, and safer marketplace paths.
• Retained the last authorized GitHub traffic sample when scheduled Actions cannot read traffic endpoints.

Verified evidence

• 37/37 regression tests pass.
• CI, Pages, and HOL marketplace scanning all pass on the tagged commit.
• The controlled low-load benchmark runs four independent tasks for 310+ seconds each: 5m 10s wall time versus 20m 40s observed serial equivalent, saving 15m 30s (4.00×).
• Benchmark wording remains explicit: serial equivalent is the sum of observed independent task runtimes; the suite is not rerun serially.

Live report: https://cloudguo123.github.io/mac-parallel-accelerator/

Mac Parallel Accelerator v0.9.2

Highlights

• Fixes immutable plan verification across Python/JavaScript MCP JSON round trips while preserving tamper detection.
• Adds an outcome-first bilingual README, 1280×640 social card, 20-second live-progress demo, and public sharing assets.
• Adds privacy-safe weekly GitHub growth metrics, community issue/discussion templates, roadmap, contribution guide, citation metadata, and launch kit.
• Adds a machine-readable real-project benchmark protocol without relabeling the controlled scheduler test as external evidence.

Verified evidence

• 35/35 regression tests pass.
• 6/6 release gates pass on local and GitHub CI.
• Controlled 5+ minute run: 20m41s serial equivalent, 5m10s parallel wall, 15m31s saved, 4.00× observed.
• Secret scan: 0 high-confidence matches.
• GitHub Actions: all third-party actions pinned to complete commit SHAs.

The controlled result measures the retained independent workload and reporting path; it is not a universal project speedup claim.