Summary
Five agentic CLI/MCP components have updates available. Versions have been bumped in pkg/constants/version_constants.go and make recompile has been run successfully against all 223 workflows.
Component
Previous
New
Status
Claude Code
2.1.133
2.1.140
Updated
Copilot CLI
1.0.43
1.0.46
Updated
Codex
0.129.0
0.130.0
Updated
GitHub MCP Server
v1.0.3
v1.0.4
Updated
Playwright Browser
v1.59.1
v1.60.0
Updated
MCP Gateway
v0.3.6
v0.3.6
No change
Playwright MCP
0.0.75
0.0.75
No change
Playwright CLI
0.1.13
0.1.13
No change
Critical Notes
Playwright 1.60.0 ships breaking changes : long-deprecated APIs removed (Locator.ariaRef(), handle option on exposeBinding, logger on connect/connectOverCDP, videosPath/videoSize context options). Browser engines bumped to Chromium 148, Firefox 150, WebKit 26.4. Risk: Medium if any gh-aw workflows or fixtures rely on the removed APIs; no known usages in this repo.
Copilot 1.0.46 auto-approves read-only gh CLI commands (list, view, status, diff). Reduces friction but slightly widens the autonomous action surface — verify expected behavior in any workflow that pins Copilot.
All other changes are additive features and bug fixes — Risk: Low for the remaining four.
Update Claude Code
Previous: 2.1.133 → New: 2.1.140 (7 patch versions)
Key Features (2.1.139)
New unified agent view across all Claude Code sessions
/goal command for cross-turn completion conditions
/scroll-speed command for mouse wheel tuning
claude plugin details subcommand showing component inventory and costs
Hook args field supports direct command execution without shell wrapper
MCP stdio servers now receive CLAUDE_PROJECT_DIR env var
Notable Fixes (2.1.140 and earlier)
Agent subagent_type matching tolerates case and separator variations
/goal command no longer hangs silently when hooks are disabled
Settings hot-reload works correctly with symlinked configs
Read tool validation handles offset parameters with whitespace padding
MCP servers no longer disappear after /clear in IDE extensions (2.1.136)
OAuth token overwrite race condition fixed (2.1.136)
View Full Changelog
2.1.140
Improved Agent tool subagent_type matching to handle case and separator variations
Updated agent color palette
Fixed /goal command hanging silently when certain hooks are disabled
Resolved settings hot-reload issues with symlinked configuration files
Fixed background service connection failures in enterprise security environments
Addressed remote managed settings retry logic for expired tokens
Fixed /loop scheduling redundant wakeups for background task polling
Resolved Windows event-loop stalling from synchronous executable checks
Fixed Read tool validation when offset parameters use whitespace padding
Terminal cursor behavior improved when focus is lost
Plugin warnings now display when default component folders are silently ignored
2.1.139
Introduced agent view feature showing all Claude Code sessions in a unified list
Added /goal command for setting completion conditions across multiple turns
Introduced /scroll-speed command
Added claude plugin details subcommand displaying component inventory and costs
Enhanced transcript navigation with keyboard shortcuts and prompt jumping
Hook args field now supports direct command execution
PostToolUse hooks gained continueOnBlock option
MCP stdio servers now receive CLAUDE_PROJECT_DIR environment variable
Compaction prompt improved to preserve sensitive user instructions
/mcp Reconnect picks up configuration edits without restart
2.1.138 — Internal fixes only
2.1.137 — Fixed extension activation failure on Windows
2.1.136
Added env var to re-enable session quality surveys via OpenTelemetry
Introduced settings.autoMode.hard_deny for unconditional auto mode blocking
Fixed MCP servers disappearing after /clear in IDE extensions
Resolved OAuth token overwrite race condition
Fixed MCP OAuth refresh token loss during concurrent server operations
Addressed API error when extended thinking produced redacted blocks
Fixed --resume session discovery with underscores in project paths
Improved image handling with downscaling and oversized cleanup
Enhanced permission prompts and dialog consistency
Impact Assessment
Risk: Low
No breaking changes detected
All changes additive or bug fixes
Package Links
Update Copilot CLI
Previous: 1.0.43 → New: 1.0.46 (3 versions)
Timeline: 2026-05-08 through 2026-05-12
Key Features
/autopilot slash command to toggle interactive/autopilot modes (1.0.45)
/fork command to fork current session into independent session (1.0.45)
Mid-input slash commands and multi-skill invocation in single messages (1.0.44)
userPromptSubmitted hooks can bypass LLM and respond directly (1.0.44)
Optional prerelease argument to copilot update and /update (1.0.44)
Read-only gh CLI commands (list, view, status, diff) auto-approved without confirmation (1.0.46)
Notable Fixes
Deprecation warnings now displayed when CLI version may lose premium model access (1.0.46)
HTTP/2 session crash (ERR_HTTP2_INVALID_SESSION) mid-turn resolved (1.0.46)
PowerShell 7+ falls back gracefully to Windows PowerShell when unavailable (1.0.45)
OpenTelemetry output aligns with GenAI semantic conventions (1.0.45)
Startup up to ~1.5s faster on terminals with limited OSC color support (1.0.45)
agentStop hook fires correctly when agent stops via task_complete (1.0.45)
Long lines in diff view wrap at terminal width instead of truncating (1.0.46)
View Full Changelog
1.0.46 (2026-05-12)
Display warning when CLI version is deprecated and premium model access may be lost
PowerShell starts correctly when pwsh installed as .NET global tool shim
Long lines in diff view wrap at terminal width instead of being truncated
Read-only gh CLI commands (list, view, status, diff, etc.) auto-approved without prompting
Sessions no longer crash mid-turn with ERR_HTTP2_INVALID_SESSION errors
1.0.45 (2026-05-11)
Add /autopilot slash command to toggle interactive/autopilot modes
Fall back to Windows PowerShell when PowerShell 7+ (pwsh) not available
OpenTelemetry output aligns with GenAI semantic conventions
Sessions with extension permission prompts can be resumed without corruption error
agentStop hook fires correctly when agent stops via task_complete
CLI starts faster on terminals with limited OSC color query support
Add /fork command
1.0.44 (2026-05-08)
Path completion in /add-dir no longer flickers
Slash commands can appear mid-input, multiple skills per message
userPromptSubmitted hooks can bypass LLM
Faster /user list and /user switch
Optional prerelease argument to copilot update
Shell aliases and rc file settings work in ! commands
Quota display correct for Free users
Permissions granted in autopilot preserved after /clear
Effort level applies correctly when switching models
Ctrl+C while permission prompt pending no longer hangs
Invalid URL entries in settings.json skipped with warning instead of crash
Impact Assessment
Risk: Low–Medium
Behavior change: read-only gh commands auto-approved in 1.0.46 (slightly broader autonomy)
No breaking flag changes detected
Verify in next workflow run: MCP tool loading and /models with PAT auth (per version_constants.go pinning note)
Package Links
Update Codex
Previous: 0.129.0 → New: 0.130.0
Released: 2026-05-08
Key Features
Notable Fixes
View Full Changelog
Chores
Added a faster Cargo profiling build profile and disabled empty doctest targets (Add a Cargo build profile for benchmarking openai/codex#21574 , Disable empty Cargo test targets openai/codex#21584 )
Hardened dependency and CI hygiene with fully qualified GitHub Action pins, Dependabot cooldown, cargo-shear upgrade ([codex] Fully qualify hash-pins in GitHub Actions openai/codex#21436 , Upgrade cargo-shear to 1.11.2 openai/codex#21547 , [codex] Apply a Dependabot cooldown of 7 days openai/codex#21599 )
Simplified internal surfaces: removed unused device-key APIs, extra skills roots, remote thread-store implementation, string-keyed MCP tool maps (device-key: clean up unused crate openai/codex#21487 , Remove skills list extra roots openai/codex#21485 , [codex] Remove remote thread store implementation openai/codex#21596 , [codex] Remove string-keyed MCP tool maps openai/codex#21454 )
Configurable OpenTelemetry trace metadata and richer review/feedback analytics (codex-otel: add configurable trace metadata openai/codex#21556 , [codex-analytics] add tool review event schema openai/codex#18747 , [codex-analytics] plumb protocol-native review timing openai/codex#21434 , [codex] add account id to feedback uploads openai/codex#21498 )
Documentation
Impact Assessment
Risk: Low
No breaking changes; new remote-control command is additive
Internal refactors (ThreadStore, MCP tool maps) should be transparent
Package Links
Update GitHub MCP Server
Previous: v1.0.3 → New: v1.0.4
Released: 2026-05-11
Key Changes
Impact Assessment
Risk: Low
All changes are bug fixes or additive enhancements
actions_run_trigger schema fix improves correctness — workflows using this tool should benefit
get_tag now handles lightweight tags correctly
Package Links
Update Playwright Browser
Previous: v1.59.1 → New: v1.60.0
Released: 2026-05-11
Breaking Changes
Locator.ariaRef() removed — use standard locator.ariaSnapshot() pipeline instead
handle option on BrowserContext.exposeBinding and Page.exposeBinding removed
logger option on BrowserType.connect and BrowserType.connectOverCDP removed — use tracing instead
Context options videosPath / videoSize removed — use recordVideo instead
Key Features
HAR recording on Tracing : tracing.startHar() / tracing.stopHar() as first-class API
Drop API : new locator.drop() simulates external drag-and-drop of files or clipboard-like data
Aria snapshots : expect(page).toMatchAriaSnapshot() now works on Page; new boxes option appends bounding boxes for AI consumption
test.abort() : new method to abort the currently running test from a fixture, hook, or route handler
New browser.on('context') event
BrowserContext now mirrors lifecycle events from pages (download, frameattached, framedetached, framenavigated, pageclose, pageload)
New description option in getByRole() for matching accessible description
New pseudo option in expect(locator).toHaveCSS() reads computed styles from ::before / ::after
New webSocketRoute.protocols() returns WebSocket subprotocols
New noDefaults option in connectOverCDP() disables Playwright's default overrides
New webError.location() mirrors consoleMessage.location()
HTML reporter accepts .zip files directly via npx playwright show-report
Trace Viewer adds pretty-print toggle for JSON / form request and response bodies
Browser Versions
Chromium 148.0.7778.96
Mozilla Firefox 150.0.2
WebKit 26.4
Impact Assessment
Risk: Medium
Breaking changes affect only deprecated APIs that have been deprecated for some time
Browser version bumps may surface site/test regressions in dependent workflows
Affects: Playwright-based workflows in gh-aw that use the browser Docker image (mcr.microsoft.com/playwright:v1.60.0)
Migration: search workflow definitions and templates for ariaRef, exposeBinding ... handle, connectOverCDP ... logger, videosPath, videoSize
Package Links
Recommendations
Priority : Merge after running smoke tests for Claude, Copilot, and Codex engines
Testing strategy :
Run smoke-claude.lock.yml, smoke-copilot.lock.yml, smoke-codex.lock.yml to validate the new CLI versions
For Playwright, run any browser-automation workflows against the new image
Confirm Copilot MCP tools.mcp loading still works and /models does not silently fail on PATs (per pinning note in version_constants.go)
Rollout : Single PR bundling all five updates is appropriate — they are independent components and risk is bounded to their respective engines
Files Changed
pkg/constants/version_constants.go — 5 version constants updated
223 *.lock.yml files regenerated via make recompile
References
Generated by CLI Version Checker · ● 8.5M · ◷
Summary
Five agentic CLI/MCP components have updates available. Versions have been bumped in
pkg/constants/version_constants.goandmake recompilehas been run successfully against all 223 workflows.Critical Notes
Locator.ariaRef(),handleoption onexposeBinding,loggeronconnect/connectOverCDP,videosPath/videoSizecontext options). Browser engines bumped to Chromium 148, Firefox 150, WebKit 26.4. Risk: Medium if any gh-aw workflows or fixtures rely on the removed APIs; no known usages in this repo.ghCLI commands (list, view, status, diff). Reduces friction but slightly widens the autonomous action surface — verify expected behavior in any workflow that pins Copilot.Update Claude Code
Key Features (2.1.139)
/goalcommand for cross-turn completion conditions/scroll-speedcommand for mouse wheel tuningclaude plugin detailssubcommand showing component inventory and costsargsfield supports direct command execution without shell wrapperCLAUDE_PROJECT_DIRenv varNotable Fixes (2.1.140 and earlier)
subagent_typematching tolerates case and separator variations/goalcommand no longer hangs silently when hooks are disabled/clearin IDE extensions (2.1.136)View Full Changelog
2.1.140
subagent_typematching to handle case and separator variations/goalcommand hanging silently when certain hooks are disabled/loopscheduling redundant wakeups for background task polling2.1.139
/goalcommand for setting completion conditions across multiple turns/scroll-speedcommandclaude plugin detailssubcommand displaying component inventory and costsargsfield now supports direct command executionPostToolUsehooks gainedcontinueOnBlockoptionCLAUDE_PROJECT_DIRenvironment variable/mcpReconnect picks up configuration edits without restart2.1.138 — Internal fixes only
2.1.137 — Fixed extension activation failure on Windows
2.1.136
settings.autoMode.hard_denyfor unconditional auto mode blocking/clearin IDE extensions--resumesession discovery with underscores in project pathsImpact Assessment
Package Links
Update Copilot CLI
Key Features
/autopilotslash command to toggle interactive/autopilot modes (1.0.45)/forkcommand to fork current session into independent session (1.0.45)userPromptSubmittedhooks can bypass LLM and respond directly (1.0.44)prereleaseargument tocopilot updateand/update(1.0.44)Notable Fixes
ERR_HTTP2_INVALID_SESSION) mid-turn resolved (1.0.46)agentStophook fires correctly when agent stops viatask_complete(1.0.45)View Full Changelog
1.0.46 (2026-05-12)
1.0.45 (2026-05-11)
1.0.44 (2026-05-08)
prereleaseargument tocopilot updateImpact Assessment
ghcommands auto-approved in 1.0.46 (slightly broader autonomy)/modelswith PAT auth (per version_constants.go pinning note)Package Links
Update Codex
Key Features
codex remote-control— new top-level command, simpler entrypoint for starting a headless, remotely controllable app-server (add top-level remote-control command openai/codex#21424)aws loginprofiles (feat: enable AWS login credentials for Bedrock auth openai/codex#21623)view_imageresolves files through selected environment for multi-environment sessions (Route view_image through selected environments openai/codex#21143)Notable Fixes
ThreadStore(Move thread name edits to ThreadStore openai/codex#21264, Route ThreadManager rollout path reads through thread store openai/codex#21265, [codex] Fix pathless thread summaries openai/codex#21266)response.processedfor v2 streams, avoidsservice_tieron API-key compact requests (Send response.processed after remote compaction v2 openai/codex#21642, Omit service_tier from remote /responses/compact requests under API auth openai/codex#21676)codex execstartup banner (Remove exec research preview banner wording openai/codex#21683)View Full Changelog
Chores
cargo-shearupgrade ([codex] Fully qualify hash-pins in GitHub Actions openai/codex#21436, Upgradecargo-shearto 1.11.2 openai/codex#21547, [codex] Apply a Dependabot cooldown of 7 days openai/codex#21599)Documentation
cargo install --locked(Ensure all mentions of cargo-install are --locked openai/codex#21592)Impact Assessment
remote-controlcommand is additivePackage Links
Update GitHub MCP Server
Key Changes
defer _meta.ui strip to per-request RegisterTools— mcp-apps fix (fix(mcp-apps): defer _meta.ui strip to per-request RegisterTools github-mcp-server#2446)ifclabel forget_metool (Add ifc label for get_me tool github-mcp-server#2432)actions_run_triggertool schema (Prevent inputs param from being stripped from actions_run_trigger tool schema github-mcp-server#2417)get_tag(Handle lightweight tags in get_tag github-mcp-server#2400)Impact Assessment
actions_run_triggerschema fix improves correctness — workflows using this tool should benefitget_tagnow handles lightweight tags correctlyPackage Links
Update Playwright Browser
Breaking Changes
Locator.ariaRef()removed — use standardlocator.ariaSnapshot()pipeline insteadhandleoption onBrowserContext.exposeBindingandPage.exposeBindingremovedloggeroption onBrowserType.connectandBrowserType.connectOverCDPremoved — use tracing insteadvideosPath/videoSizeremoved — userecordVideoinsteadKey Features
tracing.startHar()/tracing.stopHar()as first-class APIlocator.drop()simulates external drag-and-drop of files or clipboard-like dataexpect(page).toMatchAriaSnapshot()now works onPage; newboxesoption appends bounding boxes for AI consumptiontest.abort(): new method to abort the currently running test from a fixture, hook, or route handlerbrowser.on('context')eventdescriptionoption ingetByRole()for matching accessible descriptionpseudooption inexpect(locator).toHaveCSS()reads computed styles from::before/::afterwebSocketRoute.protocols()returns WebSocket subprotocolsnoDefaultsoption inconnectOverCDP()disables Playwright's default overrideswebError.location()mirrorsconsoleMessage.location().zipfiles directly vianpx playwright show-reportBrowser Versions
Impact Assessment
mcr.microsoft.com/playwright:v1.60.0)ariaRef,exposeBinding ... handle,connectOverCDP ... logger,videosPath,videoSizePackage Links
mcr.microsoft.com/playwright:v1.60.0Recommendations
smoke-claude.lock.yml,smoke-copilot.lock.yml,smoke-codex.lock.ymlto validate the new CLI versions/modelsdoes not silently fail on PATs (per pinning note in version_constants.go)Files Changed
pkg/constants/version_constants.go— 5 version constants updated*.lock.ymlfiles regenerated viamake recompileReferences