Skip to content

📊 AI CLI Tools Digest 2026-04-29 #70

Open
Open
📊 AI CLI Tools Digest 2026-04-29#70
Labels
digest-en

Description

@github-actions
github-actions
bot
opened on Apr 29, 2026

AI CLI Tools Community Digest 2026-04-29

Generated: 2026-04-29 02:06 UTC | Tools covered: 8

Cross-Tool Comparison

The user wants me to generate a cross-tool comparison report synthesizing all the individual digests I've created. Let me analyze the data from each tool's digest and create a comprehensive comparison report.

Let me organize the key data points:

Activity Comparison:

  • Claude Code: 50 issues, 6 PRs, 1 release (v2.1.122)
  • OpenAI Codex: 50 issues, 20 PRs, 4 alpha releases
  • Gemini CLI: 50 issues, 20 PRs, 4 releases (v0.40.0, v0.40.0-preview.5, v0.41.0-preview.0, v0.41.0-nightly)
  • GitHub Copilot CLI: 47 issues, 10 PRs, 2 releases (v1.0.39, v1.0.39-0)
  • Kimi Code CLI: 11 issues, 10 PRs, 1 release (v1.40.0)
  • OpenCode: 50 issues, 20 PRs, 1 release (v1.14.29)
  • Qwen Code: 31 issues, 46 PRs, 5 releases
  • Pi: 50 issues, 18 PRs, 1 release (v0.70.6)

Shared Feature Directions:

  1. MCP (Model Context Protocol) support - Claude Code, OpenAI Codex, Gemini CLI, Kimi Code CLI, OpenCode, Qwen Code, Pi

  2. Context window management / compaction - Claude Code, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Code

  3. Approval/permission modes - Claude Code, Gemini CLI, GitHub Copilot CLI, Kimi Code CLI

  4. Session management - All tools

  5. Shell/tool improvements - All tools

  6. WSL/Windows compatibility - Claude Code, Gemini CLI, OpenCode

  7. Auto mode / yolo mode - Gemini CLI, GitHub Copilot CLI, Kimi Code CLI

  8. Hot-reload for skills/extensions - Gemini CLI, Qwen Code, Kimi Code CLI

  9. Runtime identity / sidecar files - Gemini CLI, Qwen Code, OpenCode

  10. Image/vision support - Qwen Code, Pi, OpenAI Codex

Differentiation Analysis:

  • Claude Code: Anthropic's official CLI, strong on agent reliability, billing transparency
  • OpenAI Codex: Microsoft's CLI, deep VS Code integration, ACP protocol focus
  • Gemini CLI: Google's CLI, WSL2 focus, rapid iteration with preview/nightly channels
  • GitHub Copilot CLI: Microsoft's CLI, ACP protocol, slash commands, background tasks
  • Kimi Code CLI: Moonshot AI's CLI, approval system focus, session tracking
  • OpenCode: Anomaly's CLI, Effect-based architecture, mobile optimization
  • Qwen Code: Alibaba's CLI, model cost estimation, parallel skill loading
  • Pi: Mario Zecher's CLI, lightweight, XDG compliance, Cloudflare Workers AI support

Community Momentum:

  • Most active: Qwen Code (46 PRs), OpenAI Codex (20 PRs), Gemini CLI (20 PRs), OpenCode (20 PRs)
  • Rapid iteration: Gemini CLI (4 releases), Qwen Code (5 releases), OpenAI Codex (4 releases)
  • Most issues: Claude Code, OpenAI Codex, Gemini CLI, OpenCode, Pi (all 50)
  • Least issues: Kimi Code CLI (11)

Trend Signals:

  1. MCP ecosystem expansion - all major tools supporting or improving MCP
  2. Context management sophistication - compaction, cost estimation, streaming optimization
  3. Permission/approval system maturity - granular controls, configurable timeouts
  4. Platform-specific reliability - WSL2, Windows, macOS each have distinct issues
  5. Session persistence and recovery - sidecar files, runtime identity
  6. Developer experience - hot-reload, shell improvements, telemetry
  7. Multi-model routing - cost optimization, model precedence, provider flexibility

AI CLI Tools Ecosystem — Cross-Tool Comparison Report

Date: 2026-04-29
Scope: Claude Code, OpenAI Codex, Gemini CLI, GitHub Copilot CLI, Kimi Code CLI, OpenCode, Qwen Code, Pi

1. Ecosystem Overview

The AI CLI tools landscape is experiencing rapid maturation, with eight major platforms competing for developer mindshare. Anthropic's Claude Code and OpenAI's Codex maintain the highest visibility, while Google's Gemini CLI and Microsoft's GitHub Copilot CLI offer differentiated approaches through ACP protocol support and deep IDE integration. Chinese AI labs (Moonshot, Alibaba, Anomaly) have established credible alternatives with unique feature focuses—approval systems, Effect-based architectures, and mobile optimization. The ecosystem shows convergence on MCP (Model Context Protocol) as the standard extensibility mechanism, while diverging on permission models, context management strategies, and platform-specific reliability approaches. Community feedback consistently highlights billing transparency, session persistence, and WSL2/Windows compatibility as cross-cutting pain points.

2. Activity Comparison

Tool Issues (24h) PRs (24h) Releases Release Velocity Community Size Signal
Claude Code 50 6 1 (v2.1.122) Low High (50 issues, 6 PRs)
OpenAI Codex 50 20 4 (alpha series) High High (50 issues, 20 PRs)
Gemini CLI 50 20 4 (preview/nightly) Very High High (50 issues, 20 PRs)
GitHub Copilot CLI 47 10 2 (v1.0.39) Medium High (47 issues, 10 PRs)
Kimi Code CLI 11 10 1 (v1.40.0) Low Medium (11 issues, 10 PRs)
OpenCode 50 20 1 (v1.14.29) Low High (50 issues, 20 PRs)
Qwen Code 31 46 5 (mixed) Very High High (31 issues, 46 PRs)
Pi 50 18 1 (v0.70.6) Low High (50 issues, 18 PRs)

Key Observations:

  • Qwen Code leads in PR throughput (46 PRs in 24h), indicating aggressive feature development
  • Gemini CLI and OpenAI Codex maintain rapid release cadences with preview/nightly channels
  • Kimi Code CLI shows the lowest issue volume (11), suggesting either a smaller community or better issue triage
  • Claude Code has surprisingly low PR activity (6) despite high issue engagement, indicating a core-team-driven release model

3. Shared Feature Directions

The following requirements appear across multiple tool communities, indicating industry-wide engineering priorities:

Feature Direction Tools Affected Specific Needs
MCP Ecosystem Claude Code, Codex, Gemini, Kimi, OpenCode, Qwen, Pi MCP server CLI config, remote plugin caching, project-scoped MCPs, OAuth field support
Context Management Claude Code, Gemini, Copilot, OpenCode, Qwen Auto-compaction thresholds, context window display, streaming optimization, token usage visibility
Permission/Approval Systems Claude Code, Gemini, Copilot, Kimi Configurable timeouts, granular tool whitelists, allow-all toggles, scope-aware pending requests
Session Persistence All 8 tools Sidecar files (runtime.json), session archival, crash-safe writes (fsync), session title preservation
WSL2/Windows Reliability Claude Code, Gemini, OpenCode OAuth session loss, fork table exhaustion, EPIPE crashes, input handling bugs
Auto/Yolo Modes Gemini, Copilot, Kimi Unified auto-approve semantics, orthogonal AFK mode, --yolo flag consolidation
Hot-Reload Capabilities Gemini, Qwen, Kimi Runtime skill/extension/MCP/config reload without session restart
Runtime Identity Gemini, Qwen, OpenCode PID + session ID exposure for external observability and debugging
Image/Vision Support Qwen, Pi, Codex Local model image recognition, ImageContent type support, Flux model integration
Shell Tool Improvements All 8 tools Signal forwarding, heredoc parsing, PowerShell support, background task management

4. Differentiation Analysis

Tool Primary Differentiator Target User Technical Approach
Claude Code Agent reliability & billing transparency Professional developers, enterprise Mature permission system, regression-aware release process
OpenAI Codex ACP protocol & VS Code integration Microsoft ecosystem users Protocol-first design, marketplace-centric plugin model
Gemini CLI Rapid iteration & WSL2 focus Google Cloud/AI developers Preview/nightly channels, Effect-based architecture
GitHub Copilot CLI Slash commands & background tasks GitHub power users ACP-centric, /compact, /context, /usage commands
Kimi Code CLI Approval system maturity Enterprise workflows Configurable timeouts, scope-aware approvals, session tracking
OpenCode Effect-based architecture & mobile Cross-platform developers Effect v4, Drizzle SQLite, mobile touch optimization
Qwen Code Model cost estimation & parallel loading Cost-conscious developers Per-model pricing, parallel skill discovery, DeepSeek compatibility
Pi Lightweight & XDG compliance Linux/terminal enthusiasts Minimal footprint, Cloudflare Workers AI, Shiki highlighting

Strategic Positioning:

  • Claude Code differentiates on trust (billing accuracy, regression stability) rather than feature velocity
  • OpenAI Codex bets on the ACP protocol becoming a cross-tool standard
  • Gemini CLI pursues aggressive iteration to close feature gaps quickly
  • Chinese tools (Kimi, Qwen, OpenCode) focus on local model support and cost optimization for price-sensitive markets

5. Community Momentum & Maturity

High-Momentum Tools

Tool Evidence Trajectory
Qwen Code 46 PRs/24h, 5 releases, active feature development Accelerating—aggressive feature parity push
Gemini CLI 4 releases/24h, nightly channel, 20 PRs Accelerating—rapid iteration toward v0.41
OpenAI Codex 4 alpha releases, 20 PRs, ACP protocol expansion Stable—mature release cadence
OpenCode 20 PRs, Effect architecture migration, mobile focus Accelerating—architecture modernization

Mature/Stable Tools

Tool Evidence Trajectory
Claude Code Low PR volume, high issue engagement, regression-aware Stable—quality over velocity
GitHub Copilot CLI 2 releases, focused ACP/slash command work Stable—incremental improvements
Pi 1 release, XDG compliance, lightweight focus Stable—maintenance mode with polish
Kimi Code CLI 11 issues, 10 PRs, approval system focus Stable—feature consolidation

Community Health Signals

  • Issue Resolution Rate: Claude Code, Copilot CLI, and Pi show higher closed-to-open ratios, suggesting active triage
  • Contributor Diversity: OpenCode, Qwen Code, and Gemini CLI show multi-contributor PRs; Claude Code is primarily core-team driven
  • Regression Frequency: Claude Code community explicitly tracks "regression" issues, indicating maturity awareness
  • Billing Issues: Claude Code and Copilot CLI communities frequently discuss unexpected charges—trust is a differentiator

6. Trend Signals

Industry Trends from Community Feedback

1. MCP as the Universal Extensibility Layer
Every major tool is investing in MCP support—server configuration, OAuth, remote caching, and project-scoped instances. The protocol is becoming table stakes. Developers should prioritize MCP skill development.

2. Context Management Sophistication
Auto-compaction, token usage visibility, and cost estimation are no longer optional. Tools are competing on context efficiency. Expect context-aware pricing models to become standard.

3. Permission System Maturity
The "all-or-nothing" approval model is being replaced by granular, configurable, scope-aware permission systems. Enterprise adoption depends on fine-grained control.

4. Platform-Specific Reliability Gaps
WSL2, Windows, and macOS each have distinct failure modes. No tool has achieved full cross-platform parity. Developers on non-standard environments should validate critical workflows.

5. Session Persistence as a Trust Issue
Billing transparency and session durability are community pain points. Sidecar files (runtime.json) and crash-safe writes are emerging patterns. Data integrity and observability are competitive advantages.

6. Auto Mode Fragmentation
Tools are converging on auto-approve functionality but using inconsistent terminology (--yolo, --afk, --auto-approve). Standardization would improve developer experience.

7. Local Model Support
Qwen Code, Pi, and OpenCode are investing in local LLM provider extensions. This reflects demand for offline capability and cost reduction. Local model support will become a differentiator.

8. Developer Experience Convergence
Hot-reload, shell improvements, and telemetry are universal priorities. The gap between "works" and "delightful" is where tools compete. Investment in DX pays compounding returns.

Recommendations for Developers

Scenario Recommended Tool Rationale
Enterprise with billing sensitivity Claude Code Billing transparency, regression stability
Microsoft ecosystem integration OpenAI Codex ACP protocol, VS Code deep integration
Rapid iteration / bleeding edge Gemini CLI Nightly releases, aggressive feature development
Cost optimization priority Qwen Code Model cost estimation, local model support
Approval workflow requirements Kimi Code CLI Configurable timeouts, scope-aware permissions
Mobile / cross-platform OpenCode Mobile touch optimization, Effect architecture
Linux / minimal footprint Pi XDG compliance, lightweight, Cloudflare Workers AI

Report generated from 2026-04-29 community data across 8 AI CLI tools. For tool-specific details, refer to individual digests.

Per-Tool Reports

Claude Code — anthropics/claude-code

The user wants me to generate a structured English digest for the Claude Code community based on the provided GitHub data. Let me analyze the data and create a well-organized digest following the specified sections.

Let me organize the information:

Releases:

  • v2.1.122: Added ANTHROPIC_BEDROCK_SERVICE_TIER environment variable for Bedrock service tier selection, and PR URL pasting in /resume search box

Hot Issues (top 10 by comment count):

  1. #47252 - Ultraplan stream idle timeout errors (25 comments, 13 👍)
  2. #45937 - Dispatch main conversation offline despite working Cowork (21 comments, 11 👍)
  3. #49363 - Regression: malware reminder causes subagent refusals (20 comments, 21 👍)
  4. #53262 - HERMES.md in git commits causes extra billing (17 comments, 5 👍) - CLOSED
  5. #38788 - Claude Code 2.1.83+ broken on WSL1 (15 comments, 3 👍)
  6. #36465 - RPC error: useradd failed, can't start Cowork tasks (14 comments, 3 👍)
  7. #28765 - Feature: Push notifications for completed tasks (13 comments, 39 👍)
  8. #15379 - Jupyter notebook edits don't auto-refresh in VS Code (12 comments, 19 👍)
  9. #53086 - 2.1.120 regression: --resume/--continue crash (10 comments, 32 👍) - CLOSED
  10. #15762 - Feature: Smart Session Rename (9 comments, 36 👍)

Pull Requests (6 total):

  1. #54429 - fix(hookify): drop hookify.* import prefix

  2. #54424 - fix(plugin-dev): document repository as string-only

  3. #54103 - fix(commit-commands): cover all bash invocations

  4. #54094 - fix: quote $CLAUDE_PLUGIN_ROOT in plugin hook commands

  5. #54391 - docs: clarify calculation assumptions in bug reports

  6. #54134 - fix: correct MacOS to macOS in README.md

Now I'll structure this into a professional digest format.

Claude Code Community Digest

Date: 2026-04-29
Source: github.com/anthropics/claude-code

Today's Highlights

Version v2.1.122 introduces Bedrock service tier selection via the new ANTHROPIC_BEDROCK_SERVICE_TIER environment variable, giving users control over latency vs. priority routing. The community continues to grapple with a regression in v2.1.111 where malware reminders injected into every Read and Grep cause subagent refusals—a bug supposedly fixed in v2.1.92 but now back with a vengeance. Billing-related issues are also trending, with users reporting unexpected charges tied to commit message content and ultrareview timeouts.

Releases

v2.1.122 (Latest)

Change Description
Bedrock Service Tier New ANTHROPIC_BEDROCK_SERVICE_TIER env var (default, flex, or priority) sent as X-Amzn-Bedrock-Service-Tier header
PR Resume Enhancement Pasting a GitHub PR URL into the /resume search box now locates the session that created that PR

Hot Issues

1. [Regression] Malware reminder causes subagent refusals in v2.1.111

#49363 | 20 comments | 21 👍 | Open

A previously fixed bug (#47027, resolved in v2.1.92) has resurfaced. The <system-reminder> injection into every Read and Grep operation is triggering false refusals in subagents. This is a high-severity regression affecting multi-agent workflows.

2. Ultraplan stream idle timeout errors

#47252 | 25 comments | 13 👍 | Open

The research preview "ultraplan" refinement consistently fails with Stream idle timeout - partial response received. The approval UI never appears, blocking the workflow entirely. This is the most-commented issue in the last 24h.

3. Dispatch main conversation permanently offline

#45937 | 21 comments | 11 👍 | Open

On macOS, the main Dispatch conversation shows "offline" on mobile clients despite working fine from the desktop. Individual Cowork tasks function normally, indicating a specific Dispatch threading bug rather than a connectivity issue.

4. HERMES.md in git commits routes billing to extra usage

#53262 | 17 comments | 5 👍 | Closed

A case-sensitive string HERMES.md in recent commit history silently routes API requests to "extra usage" billing instead of Max plan quota. One user reported $200 in unexpected charges. Now closed—likely addressed in v2.1.122.

5. --resume / --continue crash in v2.1.120

#53086 | 10 comments | 32 👍 | Closed

CLI resume functionality crashes with "g9H is not a function" on v2.1.120. The workaround (interactive /resume from REPL) works, but this is a breaking regression for automation scripts. Closed—patch likely shipped in v2.1.122.

6. Jupyter notebook edits don't auto-refresh in VS Code

#15379 | 12 comments | 19 👍 | Open

Claude Code writes notebook changes to disk but VS Code doesn't refresh the editor view automatically. Users must manually reload. Affects the data science workflow use case significantly.

7. Auto-compact triggers at ~367K tokens on Opus 4.7

#54056 | 3 comments | 0 👍 | Open

A regression causing premature context compaction around 367K tokens on Opus 4.7 (1M context). This undermines the value of large-context models for long-running analysis tasks.

8. Sandbox blocks macOS SystemConfiguration API

#42857 | 3 comments | 0 👍 | Open

The seatbelt sandbox blocks com.apple.SystemConfiguration.configd, causing Rust CLI tools (including Codex CLI) to panic. This is a platform compatibility issue affecting developers using Rust-based tooling.

9. Claude Code execution fails 100% on Web

#54497 | 3 comments | 0 👍 | Open

The web-based Claude Code and standalone app fail 100% of the time. This is a critical blocker for web users and was reported today.

10. /doctor mislabels MCP config path

#54495 | 2 comments | 0 👍 | Open

The /doctor command in v2.1.122 reports home-scope .mcp.json parse errors using a cwd-template path, masking the actual file location. This obscures debugging for MCP configuration issues.

Key PR Progress

PR Author Summary
#54429 @Codeturion Fixes hookify import resolution by dropping the hookify.* prefix so hooks work in versioned cache
#54424 @Codeturion Documents that repository in plugin manifests must be a string (not object)
#54103 @Codeturion Adds missing git commands (git diff HEAD, git branch --show-current, git checkout -b, git switch -c) to /commit-push-pr allowed-tools
#54094 @Codeturion Quotes $CLAUDE_PLUGIN_ROOT in plugin hook commands to handle paths with spaces
#54391 @onthebed Clarifies bug report docs to request exact inputs/assumptions for calculation bugs
#54134 @sanath8 Corrects MacOS → macOS in README.md installation section

Note: All 6 PRs in the last 24h are from community contributors, primarily @Codeturion, focusing on plugin development ergonomics and documentation improvements.

Feature Request Trends

The most-requested enhancements cluster around three themes:

Theme Issues Top Request
Notifications & Async #28765 Push notifications when remote-control tasks complete while app is backgrounded (39 👍)
Session Management #15762, #34243 Smart session rename and programmatic rename from skills/commands (36 👍, 13 👍)
Cowork Reliability #54483 Deterministic VM filesystem mounts that survive Claude Desktop restart

Developer Pain Points

  1. Billing Surprises — Users are being charged unexpectedly due to commit message content (HERMES.md) and ultrareview timeouts consuming quota credits.

  2. Regression Instability — Multiple fixes are not holding across releases (malware reminder bug, auto-compact triggers, resume crashes). The pace of releases may be introducing instability.

  3. Platform Fragmentation — WSL1, Windows, and macOS each have distinct issues affecting core functionality (sandbox restrictions, session archival, Dispatch threading).

  4. MCP & Plugin DX — Path handling bugs in plugin hooks (unquoted variables, import resolution) create friction for the growing plugin ecosystem.

  5. Context Management — Premature compaction and stream timeouts undermine the value proposition of large-context models like Opus 4.7.

End of digest. Data collected 2026-04-29.

OpenAI Codex — openai/codex

The user wants me to generate an OpenAI Codex community digest for 2026-04-29 based on the provided GitHub data. Let me analyze the data and create a structured digest with the requested sections.

Let me organize the information:

Releases (last 24h):

  • rust-v0.126.0-alpha.12
  • rust-v0.126.0-alpha.9
  • rust-v0.126.0-alpha.11
  • rust-v0.126.0-alpha.10

These are alpha releases of the Rust component.

Latest Issues (top 30 by comment count):

  1. #19464 - 1M token context for GPT-5.5 (74 comments, 108 👍)

  2. #8648 - Codex replies to earlier messages (53 comments, 48 👍)

  3. #9634 - Token refresh bug (37 comments, closed)

  4. #16088 - Empty .codex file regression (32 comments, 74 👍)

  5. #2628 - Project-specific MCPs (30 comments, 141 👍, closed)

  6. #18258 - Computer Use plugin unavailable on macOS (30 comments, 35 👍)

  7. #18404 - Computer Use unavailable on Intel Mac (14 comments)

  8. #17318 - Can't change model/reasoning efforts (13 comments)

  9. #10867 - Custom model providers in app (9 comments)

  10. #8649 - Auto reasoning effort (8 comments)

  11. #11954 - Remote compact task error (7 comments)

  12. #12862 - CLI worktree/tmux flags (6 comments, 30 👍)

  13. #19793 - Security scan page fails (5 comments)

  14. #20048 - Windows Browser Use fails (4 comments)

  15. #17139 - MallocStackLogging on macOS (4 comments)

  16. #19930 - Intel macOS blurred overlay (4 comments, closed)

  17. #19011 - Windows automation thread issue (4 comments)

  18. #14330 - Environment isolation (4 comments)

  19. #18708 - Edit any message (4 comments)

  20. #14164 - Mobile companion (3 comments, closed)

  21. #19747 - Automations sidebar disappeared (3 comments)

  22. #16669 - Adjustable chat width (3 comments)

  23. #20054 - Cyrillic characters bug (3 comments, closed)

  24. #19674 - arg0 error messages lack context (3 comments)

  25. #20025 - Homebrew cask hangs (3 comments)

  26. #19364 - Invalid configuration error (3 comments, closed)

  27. #19607 - Rate limit usage (3 comments)

  28. #19921 - Expose plan-mode-prompt (3 comments)

  29. #20116 - VS Code extension UX (2 comments)

  30. #20104 - Scrolling issue (2 comments)

Latest Pull Requests (top 20 by comment count):

  1. #20058 - Environment provider snapshot
  2. #20096 - Remote installed plugin cache for skills and MCP
  3. #19843 - Remove marketplace
  4. #20120 - Remove sandbox full-auto shortcut
  5. #20119 - Add sandbox replay json mode
  6. #20118 - Add sandbox profile config controls
  7. #20117 - Add explicit sandbox permission profiles
  8. #17090 - Emit tool item events
  9. #17089 - Add tool item event schemas
  10. #18748 - Emit terminal tool review events
  11. #18747 - Add tool review event schema
  12. #20050 - Type client response payloads
  13. #17088 - Ingest server requests and responses
  14. #19840 - Persisted hook enablement state
  15. #20113 - Fix heredoc parsing file_redirect
  16. #20109 - Update Bedrock Mantle endpoint (closed)
  17. #20111 - Bound advisory system bwrap startup probe
  18. #20049 - Expose provider capability bounds (closed)
  19. #20086 - Fix plugin list workspace settings test isolation (closed)
  20. #20112 - Soften skill description budget warnings

Now I'll organize this into the requested format.

OpenAI Codex Community Digest

Date: 2026-04-29
Source: github.com/openai/codex

1. Today's Highlights

The Codex team shipped four new alpha releases (v0.126.0-alpha.9 through alpha.12) focused on Rust backend improvements. Community discussion is dominated by the GPT-5.5 context window limitation—developers are requesting 1M token support—and ongoing issues with the Computer Use plugin on macOS. Several significant sandbox and permission management PRs are under review, signaling upcoming CLI improvements.

2. Releases

Version Type Notes
rust-v0.126.0-alpha.12 Alpha Latest in the alpha series
rust-v0.126.0-alpha.11 Alpha —
rust-v0.126.0-alpha.10 Alpha —
rust-v0.126.0-alpha.9 Alpha —

All four releases were published within the last 24 hours. No changelogs were provided; these appear to be incremental Rust backend updates. Users on alpha channels should update via cargo install codex-cli or Homebrew.

3. Hot Issues

1. Support 1M token context for GPT-5.5 in Codex
74 comments · 108 👍 · OPEN
The most-discussed issue this period. Users argue that GPT-5.5's documented 400K context window is insufficient for large codebase analysis. The request aligns with OpenAI's broader context expansion roadmap. Why it matters: Affects power users working with monorepos and long documentation.

2. Codex replies to earlier messages instead of latest one
53 comments · 48 👍 · OPEN
A persistent conversation-level bug where Codex responds to outdated context rather than the most recent user message. Why it matters: Breaks multi-turn workflows and leads to confusing, irrelevant responses.

3. Token refresh failure: "refresh token was already used"
37 comments · 5 👍 · CLOSED
Users unable to maintain authenticated sessions due to token refresh race conditions. Why it matters: Impacts reliability of long-running Codex sessions.

4. Empty .codex file left behind when starting a thread without .codex
32 comments · 74 👍 · OPEN
Regression in WSL environments where Codex creates spurious .codex files. Why it matters: Pollutes repositories and may interfere with project configuration.

5. Project-specific MCPs
30 comments · 141 👍 · CLOSED
Feature request to scope MCP server configurations per project. Why it matters: Highly upvoted (141 👍); enables better multi-project workflow isolation.

6. Computer Use plugin unavailable on macOS despite bundled files
30 comments · 35 👍 · OPEN
Users report the Computer Use plugin shows as unavailable even when bundled files exist. A workaround involving features.apps = true and cache repair is documented. Why it matters: Blocks adoption of the browser automation feature.

7. Can't change model and reasoning efforts
13 comments · 19 👍 · OPEN
UI bug preventing model switching and reasoning effort adjustments in the Codex App. Why it matters: Limits user control over performance/cost tradeoffs.

8. Support custom model providers in app
9 comments · 11 👍 · OPEN
CLI supports custom model providers via /model, but the App lacks this functionality. Why it matters: Hinders users with self-hosted or third-party model endpoints.

9. CLI: add --worktree and --tmux flags for isolated sessions
6 comments · 30 👍 · OPEN
Request for first-class CLI flags to spawn Codex in git worktrees with optional tmux attachment. Why it matters: Addresses a common scripting workaround; 30 👍 indicates strong demand.

10. Rate limit usage notifications
3 comments · 0 👍 · OPEN
Plus plan users hitting usage limits with unclear upgrade prompts. Why it matters: Affects accessibility and user experience for non-Pro subscribers.

4. Key PR Progress

1. Add explicit sandbox permission profiles
Author: viyatb-oai · OPEN
Introduces explicit sandbox permission profiles, moving away from implicit config loading. Foundation for subsequent PRs (#20118, #20119).

2. Add sandbox profile config controls
Author: viyatb-oai · OPEN
Extends #20117 with granular config controls for sandbox profiles, enabling standalone testing with custom cwd.

3. Add sandbox replay json mode
Author: viyatb-oai · OPEN
Allows faithful replay of resolved sandbox state without re-reading ambient config. Completes the sandbox profile stack.

4. Remove sandbox full-auto shortcut
Author: viyatb-oai · OPEN
Cleans up legacy --full-auto shortcut now that explicit profile selection is available.

5. Use remote installed plugin cache for skills and MCP
Author: xl-openai · OPEN
Fetches and caches remote plugin state, enabling skills to load from remote-installed plugins without local marketplace entries.

6. Remove marketplace
Author: canvrno-oai · OPEN
Major refactoring to eliminate the plugin marketplace dependency.

7. Add environment provider snapshot
Author: starr-openai · OPEN
Refactors EnvironmentProvider to return concrete Environment instances, improving local/remote environment handling.

8. Emit tool item events from item lifecycle
Author: rhan-oai · OPEN
Part of a multi-PR analytics stack for tracking tool usage and review events.

9. Add persisted hook enablement state
Author: abhinav-oai · OPEN
Enables users to persist hook preferences and apply changes to open sessions.

10. Fix heredoc parsing file_redirect
Author: dylan-hurd-oai · OPEN
Patches a regression where heredocs incorrectly permitted file redirects to be approved by rules.

5. Feature Request Trends

Analysis of 50 recent issues reveals these dominant themes:

Theme Frequency Examples
Context window expansion High #19464 (1M tokens), #19607 (rate limits)
Platform-specific fixes High #18258, #18404 (macOS), #20048 (Windows)
MCP/Plugin improvements Medium-High #2628 (project-specific MCPs), #20096 (remote plugin cache)
CLI extensibility Medium #12862 (worktree/tmux), #8649 (auto reasoning)
Custom model support Medium #10867 (app integration)
UX/UI enhancements Medium #18708 (edit any message), #16669 (adjustable width)
Sandbox/security Medium #20117-#20120 (permission profiles)

6. Developer Pain Points

  • Computer Use plugin failures: Multiple reports (#18258, #18404) of the plugin showing "unavailable" on macOS, blocking browser automation workflows.
  • Token/auth issues: #9634 and #19364 highlight authentication and configuration errors disrupting sessions.
  • Platform regressions: Windows automation threads (#19011), Intel Mac UI bugs (#19930), and WSL .codex file pollution (#16088).
  • CLI startup hangs: #20025 reports Homebrew-installed Codex hanging with no output, making debugging impossible.
  • Context limitations: Repeated requests for larger context windows (#19464) and clearer rate limit messaging (#19607).
  • Missing App features: Custom model providers (#10867) and model/reasoning effort controls (#17318) unavailable in the desktop App despite CLI support.

Digest generated from GitHub data · For questions, open a discussion at github.com/openai/codex/discussions

Gemini CLI — google-gemini/gemini-cli

Let me analyze the GitHub data for the Gemini CLI repository and create a structured digest for 2026-04-29.

Let me organize the information:

Releases (last 24h):

  1. v0.41.0-preview.0 - version bump and fix for list suggestion
  2. v0.41.0-nightly.20260428.gc17400b83 - fix for custom theme text schema and manual update command
  3. v0.40.0 - version bump and OpenSSL retry fix
  4. v0.40.0-preview.5 - cherry-pick fix

Issues (50 total, showing top 30):
Key issues to highlight:

  1. #24880 - Google Sign-in 403 API Error (41 comments) - HIGHLY ACTIVE

  2. #22745 - AST-aware file reads investigation (5 comments)

  3. #22323 - Subagent recovery after MAX_TURNS (4 comments)

  4. #26111 - WSL2 critical bugs cascade (3 comments)

  5. #24916 - Permission asking repeatedly (3 comments)

  6. #26114 - Large paste input issue (2 comments)

  7. #25166 - Shell command stuck with "Waiting input" (2 comments)

  8. #23571 - Model creates tmp scripts randomly (2 comments)

  9. #26117 - WSL2 comprehensive reliability failure (1 comment)

  10. #25216 - EISDIR error on temp path (1 comment)

  11. #24915 - Thick black border UI issue (1 comment)

  12. #24246 - 400 error with >128 tools (1 comment)

  13. #24202 - SSH text scrambled (1 comment)

  14. #23582 - Subagents awareness of approval modes (1 comment)

  15. #22819 - Memory routing global vs project (1 comment)

  16. #22816 - Multiple indents for dependencies (1 comment)

  17. #22809 - Tune main agent for proactive memory writes (1 comment)

  18. #22746 - AST aware CLI tools investigation (1 comment)

  19. #22672 - Discourage destructive behavior (1 comment)

  20. #26154 - Model over-escalates network timeout (0 comments)

  21. #25218 - Table streaming rendering bug (0 comments)

  22. #24943 - Clarify parallel tool call layouts (0 comments)

  23. #24935 - Corruption after exiting editors (0 comments)

  24. #24546 - Helper to detect SSH (0 comments)

  25. #24470 - Scroll issues with long chats (0 comments)

  26. #24353 - Robust component level evaluations (0 comments)

  27. #24037 - Tracker update during replanning (0 comments)

  28. #23925 - Enable tracker by default (0 comments)

  29. #23897 - Behavioral eval for subagent rejections (0 comments)

  30. #23823 - Update utility models to 3.1 flash lite (0 comments)

Pull Requests (50 total, showing top 20):

  1. #26153 - Respect logPrompts flag for sensitive fields (OPEN)
  2. #26150 - Update documentation workflows with workspace trust (CLOSED)
  3. #26152 - Respect logPrompts flag for sensitive fields (CLOSED)
  4. #26143 - Refactor acpClient into specialized files (OPEN)
  5. #26149 - Expose runtime identity for external observers (OPEN)
  6. #25980 - Don't crash when @-mention captures non-path blob (OPEN)
  7. #25352 - Massive logs cause scrolling lag (OPEN)
  8. #26147 - Add EvalMetadata JSDoc annotations (OPEN)
  9. #26148 - Fix issue with box edges (OPEN)
  10. #25605 - Forward termination signals to child process (OPEN)
  11. #25642 - Forward termination signals to child process (CLOSED)
  12. #25260 - Scrub CI env vars in dev mode (OPEN)
  13. #25649 - Strip CI_* env vars in dev mode (CLOSED)
  14. #26136 - Disconnect extension-backed MCP clients (CLOSED)
  15. #25135 - Add /enhance command (OPEN)
  16. #26141 - Add missing oauth fields support (CLOSED)
  17. #24293 - Add missing return after 501 (CLOSED)
  18. #26088 - Add F10 fallback for approval mode cycling (OPEN)
  19. #25190 - Defense techniques fix (OPEN)
  20. #25978 - Point plan-mode session retention docs (CLOSED)

Now I'll organize this into a structured digest.

Gemini CLI Community Digest — 2026-04-29

1. Today's Highlights

The past 24 hours saw significant activity around WSL2 reliability, with multiple critical issues filed by a single user documenting cascading failures including OAuth session loss, EPIPE crashes, and fork table exhaustion. The team also merged two PRs addressing sensitive field logging when logPrompts: false, and continued rapid iteration toward v0.41.0 with nightly builds incorporating schema fixes and update command improvements.

2. Releases

Version Key Changes
v0.41.0-preview.0 Version bump to nightly; fix: only show list suggestion when partial input is empty
v0.41.0-nightly.20260428.gc17400b83 Fix: add missing response key to custom theme text schema; provide manual update command when auto-update fails
v0.40.0 Stable release; includes OpenSSL 3.x SSL error retry fix for streaming
v0.40.0-preview.5 Cherry-pick of fix 54b7586 to release branch

Full changelog: google-gemini/gemini-cli@v0.40.0-preview.4...v0.40.0-preview.5

3. Hot Issues

# Issue Why It Matters Community Reaction
#24880 Google Sign-in returns 403 "caller does not have permission" High-impact auth failure blocking users; 41 comments indicates widespread confusion 41 comments, actively triaged
#26111 WSL2: OAuth session loss, hook schema breaking change, EPIPE crashes, --yolo mode blocked Documents a cascade of WSL2-specific failures affecting production use 3 comments, flagged critical
#26117 WSL2 comprehensive reliability failure: 7 incidents, fork table exhaustion Compares Gemini CLI against Claude Sonnet 4.6; raises questions about WSL2 stability 1 comment, marked critical
#26154 Model over-escalates transient network timeout to critical deadlock Reports model incorrectly instructs wsl --shutdown for minor network hiccups Just filed, needs triage
#25166 Shell command execution stuck with "Waiting input" after completion Reproducible hang after simple commands; blocks automation workflows 2 comments, 3 👍
#24916 CLI repeatedly asks for file permissions Permission persistence not working; users must re-approve same files 3 comments, area/security
#26114 Large paste input causes premature line-break execution WSL2-specific input handling bug; paste triggers unintended execution 2 comments, just filed
#22323 Subagent reports GOAL success despite hitting MAX_TURNS Hides interruption status; can mislead users about task completion 4 comments, priority/p1
#22745 EPIC: Assess AST-aware file reads, search, and mapping Potential major improvement to tool precision and token efficiency 5 comments, maintainer-only tracking
#25218 Table renders incrementally during streaming, breaking screen reader layout Accessibility issue; tables re-render per-chunk instead of waiting for completion Just filed, area/core

4. Key PR Progress

# PR Status Summary
#26152 Respect logPrompts flag for logging sensitive fields CLOSED Gates telemetry events on getTelemetryLogPromptsEnabled() across four event classes
#26150 Update documentation workflows with workspace trust CLOSED Fixes CI failures in docs-audit.yml and release-notes.yml caused by untrusted directory errors
#26143 Refactor acpClient into specialized files OPEN Phase 1 of ACP modularization; decomposes monolithic acpClient.ts into focused modules
#26149 Expose runtime identity for external observers OPEN Adds runtime.json sidecar with pid, sessionId, workDir, hostname for external tooling
#25980 Don't crash when @-mention captures non-path blob OPEN Fixes unhandled promise rejection when JSON-like content is pasted or hallucinated
#25605 Forward termination signals to relaunched child process OPEN Ensures SIGTERM/SIGHUP/SIGINT reach child spawned by relaunchAppInChildProcess
#25135 Add /enhance command to improve user prompts OPEN New command leveraging LLM to refine and expand prompts using conversation history
#26136 Disconnect extension-backed MCP clients in stopExtension CLOSED Fixes extension-backed MCP clients remaining connected after extension unload
#26141 Add missing oauth fields support in subagent parsing CLOSED Adds issuer, audiences, redirect_uris support to match full MCPOAuthConfig schema
#26088 Add F10 fallback for approval mode cycling OPEN Improves Windows/WezTerm compatibility where Shift+Tab sequences are misparsed

5. Feature Request Trends

Analysis of open issues reveals these dominant feature directions:

  1. WSL2/Windows reliability — Multiple reports of OAuth failures, EPIPE crashes, and input handling bugs specific to WSL2 environments; users are comparing against Claude CLI stability.

  2. Memory & context management — Several issues (#22819, #22809, #22746) track work on global vs. project memory routing, proactive memory writes, and AST-aware codebase mapping.

  3. Agent robustness — Requests for better subagent awareness of approval modes (#23582), discouragement of destructive commands (#22672), and recovery from MAX_TURNS (#22323).

  4. Tool scalability — Issue #24246 flags 400 errors when >128 tools are available; suggests smarter tool scoping.

  5. UI/UX polish — Requests for scroll stability in long chats (#24470), SSH detection helper (#24546), and cleaner parallel tool call layouts (#24943).

  6. Security & permissions — Repeated permission prompts (#24916) and sensitive field logging (#26153) indicate need for clearer permission persistence and telemetry controls.

6. Developer Pain Points

The most frequently cited frustrations from the community:

  • WSL2 instability: OAuth session loss, fork table exhaustion, and input handling bugs are blocking production use cases.
  • Permission fatigue: Users report being asked to approve the same files repeatedly, suggesting permission state isn't persisting correctly.
  • Shell hangs: Commands completing but leaving the CLI stuck in "Waiting input" state, breaking automation.
  • Streaming UX issues: Tables re-rendering per-chunk, scrollbar jumping, and screen flashes during long chat sessions.
  • Signal forwarding gaps: Child processes spawned by the CLI ignore termination signals, causing orphaned processes.
  • Model misbehavior: Model over-reacting to transient issues (e.g., instructing wsl --shutdown for timeouts) and creating tmp scripts in random directories.

Digest generated from github.com/google-gemini/gemini-cli activity on 2026-04-29.

GitHub Copilot CLI — github/copilot-cli

The user wants me to generate a GitHub Copilot CLI community digest for 2026-04-29 based on the provided GitHub data. Let me analyze the data and create a structured digest with the requested sections.

Let me organize the information:

Releases:

  • v1.0.39 (2026-04-28): ACP permission mode toggle, new slash commands (/compact, /context, /usage, /env), ctrl+x → b to move tasks to background, transient pipe errors fix
  • v1.0.39-0: Same ctrl+x → b feature

Issues (top 30 by comments):

  1. #2725 [CLOSED] - GPT-5.4 model picker hides Extra High - 33 comments

  2. #2591 [CLOSED] - Infinite premium requests consumed per tool invocation - 32 comments

  3. #1044 [OPEN] - Slash commands support in copilot --acp - 12 comments

  4. #1455 [CLOSED] - Auto-inject Co-authored by Copilot - 9 comments

  5. #2205 [OPEN] - Scroll in terminal usability issue - 9 comments

  6. #1973 [OPEN] - Tool whitelist for Interactive Mode - 8 comments

  7. #411 [CLOSED] - Support Windows PowerShell 5.1 - 7 comments

  8. #1928 [OPEN] - Allow to pause copilot work - 6 comments

  9. #334 [OPEN] - Add shell completions - 6 comments

  10. #2408 [OPEN] - Issues auto-updating CLI - 6 comments

  11. #2282 [OPEN] - MCP server connection issues - 6 comments

  12. #1245 [CLOSED] - ACP add /compact slash command - 5 comments

  13. #1262 [CLOSED] - ACP return additional model metadata - 4 comments

  14. #1244 [CLOSED] - ACP add /usage slash command - 4 comments

  15. #2643 [OPEN] - preToolUse silent command rewrite - 4 comments

  16. #893 [CLOSED] - Instant compaction on simple prompt - 4 comments

  17. #2630 [OPEN] - Custom agent MCP servers not connected - 4 comments

  18. #1464 [OPEN] - Skills beyond alphabetical position unreachable - 4 comments

  19. #2314 [OPEN] - Skills prompt injection truncates skills - 3 comments

  20. #1392 [OPEN] - OmniSharp LSP server timeout - 2 comments

  21. #2967 [OPEN] - Opus 4.7 small context window - 2 comments

  22. #2720 [OPEN] - Speculative context for sub-agents - 2 comments

  23. #1851 [OPEN] - Show context window size in /model - 2 comments

  24. #2795 [OPEN] - --agent with --plugin-dir not working - 2 comments

  25. #1091 [CLOSED] - Add MCP server from command line - 1 comment

  26. #2943 [OPEN] - openrouter integration - 1 comment

  27. #2174 [OPEN] - Custom tool whitelist - 1 comment

  28. #2052 [OPEN] - Persistent Token/Context Usage Indicator - 1 comment

  29. #1688 [OPEN] - Configurable auto-compaction threshold - 1 comment

  30. #1131 [OPEN] - Load user shell configuration files - 1 comment

Two pull requests were merged: one updating the README and another introducing a devcontainer.json file. Now I'll structure the digest to highlight the most impactful changes and community feedback.

GitHub Copilot CLI Community Digest

Date: 2026-04-29
Repository: github.com/github/copilot-cli

1. Today's Highlights

GitHub Copilot CLI v1.0.39 shipped with significant ACP enhancements, including new slash commands (/compact, /context, /usage, /env) and a permission toggle for allow-all mode. The release also introduces background task handling via Ctrl+X → B. Community activity remains high with 47 issues updated in the last 24 hours, including the resolution of two critical bugs: the GPT-5.4 model picker hiding Extra High tier and excessive premium request consumption during tool invocations.

2. Releases

v1.0.39 (2026-04-28)

Release: https://github.com/github/copilot-cli/releases/tag/v1.0.39

Category Changes
ACP Sessions Allow ACP clients to toggle allow-all permission mode via session configuration
New Slash Commands Added /compact, /context, /usage, and /env commands for ACP sessions
Background Tasks Press Ctrl+X → B to move the current running task or shell command to the background
Bug Fixes Fixed transient pipe errors on child process stdio

v1.0.39-0 (2026-04-28)

Release: https://github.com/github/copilot-cli/releases/tag/v1.0.39-0

  • Backported the Ctrl+X → B background task feature as a standalone improvement

3. Hot Issues

1. GPT-5.4 Model Picker Hides Extra High Tier [CLOSED]

Issue: github/copilot-cli#2725 | 33 comments | 21 👍

The /model picker for GPT-5.4 only displays Low/Medium/High effort levels, even though xhigh still functions at runtime. This UI inconsistency frustrated users who expected full model capabilities. Status: Resolved—likely addressed in recent releases.

2. Infinite Premium Requests Consumed Per Tool Invocation [CLOSED]

Issue: github/copilot-cli#2591 | 32 comments | 13 👍

A critical billing bug caused each tool invocation or thinking step to consume additional premium requests, with users reporting 80-100 requests per single action. Impact: Significant unexpected costs for Pro+ subscribers. Status: Closed—fix deployed.

3. Slash Commands Support in copilot --acp [OPEN]

Issue: github/copilot-cli#1044 | 12 comments | 0 👍

ACP frontend cannot use slash commands because the protocol lacks available_commands_update. This blocks integration with editors like Zed. Why it matters: Limits ACP adoption across the ecosystem.

4. Scroll Navigation Regression in Terminal [OPEN]

Issue: github/copilot-cli#2205 | 9 comments | 7 👍

Mouse scroll now navigates input history instead of agent output history—a usability regression since the last version. Impact: Degrades terminal workflow for agent interaction.

5. Tool Whitelist for Interactive Mode [OPEN]

Issue: github/copilot-cli#1973 | 8 comments | 12 👍

Users request granular control over which tools auto-approve in interactive mode. Currently, it's all-or-nothing (/allow-all). Community demand: High (12 👍) for safer read-only operation approvals.

6. Support Windows PowerShell 5.1 [CLOSED]

Issue: github/copilot-cli#411 | 7 comments | 5 👍

CLI required PowerShell 6+ but Windows 11 ships with 5.1, blocking default installations. Status: Closed—compatibility expanded.

7. Allow Pause During Copilot Work [OPEN]

Issue: github/copilot-cli#1928 | 6 comments | 1 👍

No mechanism exists to pause an agent mid-task and provide additional instructions. Users want to steer direction without restarting sessions.

8. Shell Completions Missing [OPEN]

Issue: github/copilot-cli#334 | 6 comments | 11 👍

CLI lacks copilot completion <shell> for tab completions in zsh/bash/fish. Community demand: Moderate-to-high interest for developer experience improvement.

9. Auto-Update Issues [OPEN]

Issue: github/copilot-cli#2408 | 6 comments | 3 👍

Users report requiring multiple sessions to receive CLI updates (versions 1.0.11 → 1.0.14). Impact: Delayed access to bug fixes and features.

10. MCP Server Connection Failures [OPEN]

Issue: github/copilot-cli#2282 | 6 comments | 1 👍

Windows users via WinGet cannot connect to MCP servers, receiving "Failed to connect to MCP server 'github-mcp-server'" errors. Impact: Blocks MCP ecosystem integration.

4. Key PR Progress

PR Title Status Summary
#3018 Update README.md CLOSED Documentation update with CCPA compliance checklist
#2970 Create devcontainer.json CLOSED Added development container configuration for contributors
#1245 ACP: add /compact slash command CLOSED Implemented conversation compaction command for ACP sessions
#1262 ACP: return additional model metadata CLOSED Added usage multipliers and enablement data to model info responses
#1244 ACP: add /usage slash command CLOSED Implemented usage tracking command for ACP sessions
#1455 Auto-inject Co-authored by Copilot CLOSED Feature to attribute AI-assisted commits (similar to Claude)
#1091 Add MCP server from command line CLOSED CLI-based MCP server management without entering interactive session
#411 Support Windows PowerShell 5.1 CLOSED Expanded Windows compatibility to include PS 5.1
#893 Instant compaction on simple 'copilot set' CLOSED Fixed premature compaction triggers on basic commands
#2591 Infinite premium requests fix CLOSED Resolved billing bug consuming excessive requests

5. Feature Request Trends

Analysis of 47 updated issues reveals these dominant themes:

Theme Frequency Examples
Context Management High Auto-compaction thresholds (#1688), context window display (#1851), token usage indicator (#2052), Opus 4.7 context issues (#2967)
MCP Ecosystem High MCP server CLI addition (#1091), connection failures (#2282), custom agent MCP connections (#2630)
Permission/Tool Control Medium-High Tool whitelist (#1973, #2174), silent command rewrite (#2643), allow-all toggle
Model Configuration Medium Model picker fixes (#2725), OpenRouter integration (#2943), metadata display (#1262)
Shell Integration Medium Shell completions (#334), config file loading (#1131), PowerShell 5.1 support (#411)
ACP/Protocol Medium Slash commands (#1044), session controls (#1928), remote status hints
Installation/Updates Low-Medium Auto-update issues (#2408), skill reachability (#1464)

6. Developer Pain Points

  1. Billing Transparency: The infinite premium request bug (#2591) exposed a critical trust issue—users cannot verify request consumption, leading to unexpected costs.

  2. Context Window Opacity: Developers lack visibility into token usage (#2052) and context thresholds, causing performance degradation without warning.

  3. MCP Integration Friction: Connection failures (#2282) and custom agent MCP support (#2630) block adoption of the extended tool ecosystem.

  4. Interactive Mode Friction: All-or-nothing tool permissions force developers to either approve every action manually or risk destructive operations.

  5. Terminal Usability Regressions: Scroll behavior changes (#2205) and missing shell completions (#334) degrade daily workflow quality.

  6. ACP Protocol Gaps: Missing slash command support (#1044) limits editor integrations and cross-platform adoption.

  7. Update Reliability: Auto-update failures (#2408) delay access to fixes and create version fragmentation in teams.

Digest generated from GitHub Copilot CLI community data (2026-04-29)

Kimi Code CLI — MoonshotAI/kimi-cli

Let me analyze the GitHub data for Kimi Code CLI and generate a structured digest for 2026-04-29.

Key data points:

  1. Latest Release: 1.40.0 (with fixes for shell prompt and auth OAuth flows)
  2. 11 issues total, 10 shown
  3. 21 PRs total, 20 shown

Let me organize this into the requested sections:

Today's Highlights:

  • Release 1.40.0 with OAuth recovery fix and shell prompt improvements
  • Multiple approval-related improvements (timeout configurable, scope fixes)
  • Session persistence and telemetry improvements

Releases:

  • 1.40.0: OAuth flow recovery, shell prompt status bar fix

Hot Issues (10):

  1. #1823 - Configurable Approval Timeout (closed, implemented)
  2. #2040 - VS Code notification for approval requests
  3. #2111 - Too many open files crash
  4. #2107 - Window switching focus issue
  5. #2101 - Busy session metadata edits
  6. #2106 - Windows uv installation slow startup
  7. #2105 - Unified Auto Mode feature request
  8. #2103 - Sub-agent timeout too short
  9. #1971 - 401 Auth error with TUN mode (closed)
  10. #2096 - MCP tool list too long

Key PR Progress (10):

  1. #2082 - Runtime identity exposure (pid + session id)

  2. #1935 - updatedInput for command rewriting

  3. #1960 - RalphFlow architecture

  4. #2110 - /prompt-color command

  5. #2109 - Read-only mode

  6. #884 - Dependabot ruff update

  7. #2108 - Release bump to 1.40.0

  8. #2045 - Yolo mode fix + AFK mode

  9. #2102 - Preserve manual session titles

  10. #2104 - Tool media preview visibility

Feature Request Trends:

  • Approval system improvements (timeout, notifications, scope)
  • Auto mode unification
  • Read-only mode
  • Session persistence with fsync
  • Sub-agent timeout configuration
  • MCP tool list handling

Developer Pain Points:

  • Approval timeout issues
  • OAuth flow reliability
  • Windows performance issues
  • Session persistence reliability
  • Sub-agent timeout constraints
  • MCP tool list limitations

Kimi Code CLI Community Digest

Date: 2026-04-29
Source: github.com/MoonshotAI/kimi-cli

1. Today's Highlights

Kimi Code CLI v1.40.0 shipped with critical OAuth recovery improvements and shell UI enhancements. The community saw significant activity around approval system refinements, with both a configurable timeout feature landing and multiple related issues being addressed. New features for session tracking, read-only mode, and skill reloading demonstrate expanding use-case support.

2. Releases

v1.40.0

Released: 2026-04-28 | PR #2108

Key changes:

  • Shell: Active agent task count now displayed in prompt status bar (#2041)
  • Auth: OAuth flows now recover gracefully after transient failures (#2060)
  • Approval: Scoped pending requests to turn lifecycle; foreground approvals no longer auto-reject after 5 minutes (#2087)
  • Config: Default max_steps_per_turn increased from 500 to 1000 (#2088)

3. Hot Issues

# Title Author Status Why It Matters Reactions
#1823 Configurable Approval Request Timeout @alonsoguillen CLOSED Hardcoded 5-minute timeout caused workflow interruptions; now configurable with timeout_s=0 for unlimited (PR #1837) 👍 2, 💬 5
#2040 VS Code notification when approval required @taitoojp OPEN Approval dialogs hidden in minimized VS Code cause missed responses; needs native OS notifications 👍 0, 💬 4
#2111 Too many open files crashes agent @AriaShishegaran OPEN Critical stability issue on macOS; agent crashes with file descriptor exhaustion 👍 0, 💬 0
#2107 Window switching prints focus events in CLI input @CarmChao OPEN i3wm users experience garbage input when switching windows; affects Linux desktop environments 👍 0, 💬 0
#2101 Support busy session metadata edits safely @wbxl2000 OPEN Web UI blocks all edits during active sessions; need safe state-only modifications (PR #2102) 👍 0, 💬 0
#2106 Windows uv installation starts too slowly @Greenplumwine OPEN 1+ minute startup time on Windows 11 Enterprise; significant UX regression 👍 0, 💬 0
#2105 Unified Auto Mode as first-class feature @yysu OPEN Scattered auto flags (--yolo, --print) need consolidation into coherent Auto Mode concept 👍 1, 💬 0
#2103 Allow longer timeout for sub-agent @ZDGggg817 OPEN Sub-agents timeout prematurely on Windows; limits complex task decomposition 👍 0, 💬 0
#1971 401 Authentication Error when TUN Mode Enabled @ayokaa CLOSED OAuth token refresh bug in TUN mode; fixed in #2004 👍 0, 💬 0
#2093 Session persistence lacks fsync, data loss risk @linsenwang OPEN context.jsonl/wire.jsonl writes may not survive abnormal termination; needs fsync() for durability 👍 0, 💬 0

4. Key PR Progress

# Title Author Status Impact
#2082 Expose runtime identity (pid + session id) @yeelam-gordon OPEN Enables external process monitoring and debugging
#1935 Support updatedInput for command rewriting @zoorpha CLOSED Transparent input transformation via PreToolUse hooks
#1960 RalphFlow architecture with ephemeral context @ORDL-AMF OPEN Prevents infinite loops; isolates flow iterations in temp contexts
#2110 Add /prompt-color command @Mr-Nothing-L OPEN Custom user input echo color for better visual separation
#2109 Read-only mode @jie023 OPEN CLI parameter --readonly, /readonly and /execute commands for safe exploration
#2045 Fix yolo mode; add orthogonal AFK mode @RealKai42 CLOSED Separated auto-approve (yolo) from non-interactive (afk) semantics
#2098 Implement session tracking and telemetry @RealKai42 CLOSED session_started events with Wire client info; local debug server
#2097 Add /reload-skills command @cyliu0 OPEN Hot-reload skills without session restart
#2100 Serve JS assets with module-safe MIME type @Sisyphbaous-DT-Project OPEN Fixes Web UI startup failure on Windows
#2095 Auto approval with config granularity @suJayhh OPEN Fine-grained approval control per operation type

5. Feature Request Trends

Analysis of 11 recent issues reveals these dominant themes:

  1. Approval System Enhancements (3 issues)
    • Configurable timeouts, VS Code native notifications, scope improvements
    • Community wants more control and visibility over

⚠️ 内容超过 GitHub Issue 上限，完整报告见提交的 Markdown 文件。

Metadata

Metadata

Assignees

No one assigned

    Labels

    digest-en

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions