Claudex v1.6.4

Fixed

• Restored the verified source bootstrap after the GitHub repository moved to
  BeamoTech by resolving releases from the canonical repository on Unix and
  Windows.

Claudex v1.6.3

Changed

• Reduced redundant pull request verification and repeated legacy Linux setup
  while retaining native platform, security, and release checks.

Fixed

• Stabilized asynchronous status refresh and Windows self update lock test
  fixtures on slower CI hosts.

Claudex v1.6.2

Added

• Added collision safe shortcuts for single skill plugins whose plugin and
  skill names match, while retaining the canonical namespaced invocation.
• Added skill diagnostics for source plugin MCP, app, hook, agent, language
  server, and settings components that remain on their native harness route.

Fixed

• Stopped a native Claude plugin installed for one project from renaming an
  imported Codex plugin in every other project, and honored disabled native
  plugins when reserving namespaces.
• Made the PowerShell verification entrypoint run reliably under Windows
  PowerShell 5.1 and PowerShell 7.6 on macOS with platform specific process
  fixtures, isolated Unix state, and deterministic cache assertions.
• Made the PowerShell self update race suite portable across Windows and macOS,
  and added it to the macOS CI leg so both filesystem identity paths stay covered.
• Preserved the conventional signal exit status when an interactive package
  launcher traps a foreground group signal and exits before its wrapper callback.

Claudex v1.6.1

Fixed

• Prevented account changes that overlap watcher startup or synchronization
  from becoming an unsynchronized baseline on Unix or Windows.
• Made usage refresh contention reject fresh ownerless locks promptly without
  preparing publication files or aging the lock into its own recovery window.
• Gave the complete Unix release suite enough hosted runner time to finish and
  enforced the workflow budget for both LF and CRLF checkouts.

Claudex v1.5.8

Fixed

• Preserved the native Claude Code -- delimiter through the PowerShell host
  process and exercised it through the installed launcher's process boundary.
• Preferred sibling PowerShell shims for Windows Claude and Codex launchers so
  long managed arguments do not hit the batch command length ceiling.
• Kept empty Windows launcher argument collections as arrays so proxy recovery
  and zero argument entry points remain valid under strict mode.

Claudex v1.5.3

Added

• Added code ownership, maintainer and roadmap documentation, a dedicated
  documentation issue form, component-aware pull-request labeling, Dependabot,
  dependency review, and CodeQL scanning for a clearer and safer contributor
  workflow.
• Added public triage targets, label guidance, contribution discovery links,
  responsible AI-assisted contribution expectations, security response targets,
  and a good-faith research safe-harbor policy.

Changed

• Expanded repository governance, support expectations, contributor recognition,
  and maintainer release policy while keeping the project lightweight and
  volunteer-friendly.

Security

• Replaced ambiguous YAML plain-scalar comment stripping with bounded linear
  parsing so a large imported skill cannot stall startup with polynomial regex
  backtracking.

Claudex v1.5.1

Added

• Added transactional, retrying source installers with automatic Node.js 22,
  Claude Code, and Codex CLI provisioning, explicit Codex browser login, safe
  crash recovery, and rollback on interrupted upgrades across Unix and Windows.
• Added deterministic, type-safe release archives, an immutable draft-first
  publication workflow, full cross-platform tag gates, and automatic live
  website-installer verification after every successful release.
• Added generation-fenced package bootstrap locking so simultaneous first runs
  share one verified result without stale-lock takeovers or retry storms.

Changed

• Made auto mode honor explicit, scoped user authorization after a soft policy
  denial, ask fewer questions, stay in execution mode for routine work, and
  keep all leader, classifier, and subagent paths on Codex/OpenAI models.
• Hardened Claude Code and Codex skill compatibility with complete-file secret
  screening, policy-bound last-known-good generations, structural self-heal,
  collision-safe aliases, bounded garbage collection, and fresh diagnostics.
• Added Fable model remapping and consistent Terra (high) / Luna (medium)
  activity labels on both Bash and PowerShell launchers.
• Restricted remote proxy use to explicit HTTPS opt-in while preserving
  automatic recovery for Claudex's managed local proxy.

Fixed

• Fixed intermittent ConnectionRefused and repeated API warning loops by
  recovering stale local authentication once, cleaning failed proxy children,
  and avoiding unsafe retries against user-managed or remote proxies.
• Fixed Codex usage refresh hangs, orphaned app-server processes, unbounded
  output drains, refresh-lock races, stale cache buildup, and noisy /dev/tty
  errors; plan labels and cached usage now degrade cleanly under failure.
• Fixed split terminal writes that could leave API Usage Billing, /opus 1m,
  duplicated status lines, or broken bottom-of-screen UI visible.
• Fixed launch from a deleted working directory, ignored CLAUDEX_NODE_BIN,
  shell metacharacter handling on Windows, and self-update shell edge cases.
• Fixed package setup failure propagation, accidental npm publication risk,
  generated-release documentation scanning, and inaccurate WinGet availability
  claims.

Claudex v1.5.0

Claudex v1.5.0 adds safe, automatic compatibility for the Claude Code and Codex skills people already have installed.

Highlights:

• Discovers personal, project, admin, legacy-command, and enabled-plugin skills from both Claude Code and Codex.
• Supports normal /skill invocation plus explicit Codex-style $skill references inside Claudex.
• Preserves skill visibility, disable policy, plugin namespaces, support files, and deterministic collision aliases.
• Keeps imported skill model pins on managed Codex/OpenAI models, including [1m] selectors.
• Adds claudex skills diagnostics and matching Bash/PowerShell bridge behavior.
• Snapshots imported content into bounded immutable generations with path, credential, private-key, and special-file protections.
• Adds last-known-good recovery for interrupted refreshes.
• Improves Node 18+ migration on older Linux systems and fixes atomic updater state writes on Windows PowerShell 5.1.

The full macOS, Ubuntu, Ubuntu 20.04, Windows, package-artifact, skill-contract, security, updater, and launcher suites passed on the release commit.

Claudex v1.4.5

Fixed

• The account-bound ChatGPT plan now renders correctly in the real Claude Code/Bun welcome screen when its alternate-screen prelude and positioned banner arrive in separate writes.
• The one-shot plan filter remains byte-safe, TTY-only, geometry-scoped, callback/backpressure-preserving, and restores native output immediately after the welcome field.
• Distribution documentation now accurately lists Homebrew and Scoop as live, WinGet as pending review, and npm as not yet published.

Verified with a real pseudo-terminal capture showing ChatGPT Pro, plus green Ubuntu, macOS, Windows PowerShell, package, installer, updater, authentication, and terminal regression gates.

Claudex v1.4.4

Changed

• Managed subagent activity names now include their configured reasoning effort: Terra (high) and Luna (medium).

Fixed

• The Claude Code welcome banner now shows the detected ChatGPT subscription or workspace tier instead of the misleading API Usage Billing label.
• The welcome compatibility shim is strictly one-shot, TTY-only, byte-safe, and restores native output immediately after the atomic startup frame.
• Missing, stale, malformed, or account-switched usage caches refresh once so the displayed plan stays current.

All Ubuntu, macOS, Windows PowerShell, npm-package, updater, installer, authentication, and terminal regression gates passed.