📊 AI CLI Tools Digest 2026-03-25

[github-actions]

AI CLI Tools Community Digest 2026-03-25

Generated: 2026-03-25 00:09 UTC | Tools covered: 7

• Claude Code
• OpenAI Codex
• Gemini CLI
• GitHub Copilot CLI
• Kimi Code CLI
• OpenCode
• Qwen Code
• Claude Code Skills

---

Cross-Tool Comparison

AI CLI Tools Cross-Community Analysis — March 25, 2026

1. Ecosystem Overview

The AI CLI tooling landscape has matured into a competitive multi-polar ecosystem with six actively developed alternatives. Claude Code and OpenAI Codex lead in enterprise mindshare, while Gemini CLI, Kimi Code, Qwen Code, and OpenCode pursue differentiation through architectural bets (Effect-based reactivity, ACP protocol adoption, multi-registry orchestration). A defining pattern is the shift from "chat-first" to "agent-first" design—persistent memory, subagent hierarchies, and tool orchestration now dominate roadmaps over raw LLM access. The community is increasingly vocal about cross-platform reliability, with Windows stability and IDE parity emerging as universal pain points.

---

2. Activity Comparison

Tool	Issues (24h)	PRs (24h)	Releases (24h)	Release Velocity
Claude Code	10+ hot issues tracked	10+ community plugins submitted	None	Stable; community filling gaps
OpenAI Codex	10 hot issues	10 active PRs	4 alphas (v0.117.0-alpha.11–14)	🔥 Rapid iteration (Rust core stabilization)
Gemini CLI	10 hot issues	10 PRs merged/open	v0.36.0-preview.0	Steady preview cadence
GitHub Copilot CLI	10 hot issues	None updated	None	Stagnant; policy/MCP bugs accumulating
Kimi Code	8 issues (2 critical regressions)	7 PRs (2 merged)	None	Post-v1.25.0 stabilization
OpenCode	10 hot issues	10 PRs	v1.3.2 (with heap debugging)	Active; memory leak crisis response
Qwen Code	10 hot issues	10 PRs	v0.13.0-nightly	Nightly cadence; ACP integration focus

Note: "Issues" and "PRs" counts reflect highlighted items in community digests, not exhaustive totals.

---

3. Shared Feature Directions

Requirement	Tools	Specific Evidence
Persistent Memory & Context Bridging	Claude Code (#27140, #37159), Gemini CLI (#23724), OpenCode (#11819, #15351), Qwen Code (#2335)	Project-level storage replacing /tmp, MEMORY.md limits, session continuity across devices
Subagent Orchestration & Tool Filtering	Gemini CLI (v0.36.0 multi-registry), Claude Code (#38045 lean-agents), OpenAI Codex (implicit in scheduling)	Constrained toolsets for subagents, steering (temperature/persona), approval mode awareness
IDE Extension Parity	Kimi Code (#1334, #1560), OpenCode (#11176), Qwen Code (#1370, #2629), GitHub Copilot CLI (inherent)	VS Code/JetBrains settings sync, clipboard image paste, behavior consistency with CLI
Scheduling & Automation (/loop, daemon mode)	OpenAI Codex (#15679 /loop, #8317 time-based), Qwen Code (#2643 /loop), Gemini CLI (SDD DAG tasks)	Shift from interactive to autonomous operation
Windows Platform Hardening	Claude Code (#35710 BSOD fix, #27897 Cowork), OpenAI Codex (#15277 sandbox, #13905), OpenCode (#16069 PowerShell), Qwen Code (#2645 Git Bash)	Parallel fs ops, sandbox expansion, shell detection, path separator handling
MCP Ecosystem Maturity	Claude Code (#3426, #27263), OpenAI Codex (#15169 approval UX), Kimi Code (#1380 ACP migration)	Tool exposure reliability, external URL whitelisting, approval persistence
Lifecycle Hooks & Observability	Claude Code (plugin surge), Kimi Code (#1561 Wire 1.7 hooks), OpenCode (#6629 OpenTelemetry), OpenAI Codex (#15531 PostToolUse)	Audit logging, compliance automation, telemetry integration

---

4. Differentiation Analysis

Tool	Core Bet	Target User	Technical Distinction
Claude Code	Plugin ecosystem as extensibility layer	Enterprise teams, Cowork users	Community-driven gap-filling; Cowork containerized environments; MCP-first
OpenAI Codex	Rust core + TUI sophistication	Performance-sensitive developers, automation engineers	Rapid alpha iteration; /btw ephemeral threads, /loop scheduling; guardian review system
Gemini CLI	Multi-agent orchestration & SDD (Structured Development)	Google Cloud ecosystem, plan-mode users	Multi-registry architecture; DAG-based task planning; GSoC community scaling
GitHub Copilot CLI	Policy integration with GitHub/Copilot ecosystem	Existing Copilot subscribers	Deep GitHub org policy integration; weakest standalone differentiation
Kimi Code	ACP (Agent Client Protocol) standardization	IDE-integrated workflows, JetBrains users	ACP-native; lifecycle hooks parity with Claude Code; v1.25.0 plugin system
OpenCode	Effect-based functional reactivity	TypeScript ecosystem, desktop-first users	Effect architecture for state management; in-app browser; WebRTC P2P remote exploration
Qwen Code	Multi-platform deployment (Telegram, cmux, ACP)	Chinese-speaking developers, messaging-integrated workflows	Telegram/OSC channel integration; CJK localization priority; /loop for IoT/scheduled tasks

---

5. Community Momentum & Maturity

Tier	Tools	Indicators
🔥 Highest Momentum	OpenAI Codex, Claude Code	4 alphas/24h (Codex); 10+ community plugins/24h (Claude); sustained high engagement; enterprise-critical issues
🟢 Active & Growing	Gemini CLI, OpenCode, Qwen Code	Regular preview/nightly releases; GSoC coordination (Gemini); 20+ contributor mobilization (OpenCode memory leaks); ACP/Telegram expansion (Qwen)
🟡 Stabilizing/Recovering	Kimi Code	Post-v1.25.0 regression management; plugin system documentation debt; IDE parity catch-up
🔴 Concerning Stagnation	GitHub Copilot CLI	Zero PR activity; MCP policy drift unaddressed; model availability lag vs. VS Code; terminal regressions accumulating

Maturity Signals:

• Claude Code: Mature plugin API, but core platform gaps (ARM, FreeBSD) force community workarounds
• OpenAI Codex: Pre-release volatility with Rust rewrite; TUI sophistication outpacing stability
• Gemini CLI: Most systematic quality engineering (nightly eval runs 3→10, behavioral evals)
• OpenCode: Architectural ambition (Effect) with operational growing pains (memory leaks, auth fragility)

---

6. Trend Signals

Trend	Evidence	Developer Implication
Agent-First > Chat-First	/loop commands, scheduling PRs, persistent memory systems across all tools	Design workflows assuming autonomous execution, not just interactive assistance
Cross-Platform as Table Stakes	Windows BSOD fixes, Git Bash/MSYS2 detection, Termux/Android requests	Windows and mobile/remote environments can no longer be tier-2; test accordingly
MCP/ACP as Interoperability Layer	Claude Code MCP pain points, Kimi Code ACP migration, Qwen Code ACP expansion	Invest in protocol-compliant tool servers; expect vendor-neutral tooling standards
Community Plugin Ecosystems	Claude Code's 10+ plugin PRs in 24h, Kimi Code's beta marking	Official maintainers ceding extensibility to community; plugin quality/ security governance gaps emerging
Memory & Context Engineering	1M token window demand, compaction providers, context bridging	Shift from "prompt engineering" to "context architecture"; budget for retrieval and summarization systems
Remote & Multi-Modal Access	WebRTC P2P (OpenCode), Telegram integration (Qwen), Cowork remote access	CLI tools becoming distributed systems; consider latency, sync, and offline resilience
Enterprise Policy & Compliance	Guardian review (Codex), Policy Engine requests (Qwen), OAuth whitelisting (Claude)	Security teams will demand audit trails, approval workflows, and configuration-as-code

---

Analysis compiled from community digest data dated 2026-03-25. For real-time tracking, monitor release channels and GitHub Discussions for each project.

---

Per-Tool Reports

Claude Code — anthropics/claude-code

Claude Code Skills Highlights

Source: anthropics/skills

Claude Code Skills Community Highlights Report

Data as of 2026-03-25 | Source: github.com/anthropics/skills

---

1. Top Skills Ranking (Most-Discussed PRs)

Rank	Skill	Author	Status	Description	Discussion Highlights
1	document-typography	PGTBoos	🟡 Open	Quality control for AI-generated documents: prevents orphan words, widow paragraphs, and numbering misalignment	Addresses universal pain point affecting "every document Claude generates"; zero thumbs but high conceptual engagement
2	frontend-design (improved)	justinwetch	🟡 Open	Revised for clarity and single-conversation actionability	Focus on making instructions actually executable within token limits
3	skill-quality-analyzer + skill-security-analyzer	eovidiu	🟡 Open	Meta-skills for evaluating Skill quality across 5 dimensions and security auditing	First systematic approach to Skill validation; addresses governance gap
4	system documentation suite	TylerALofall	🟡 Open	Comprehensive evidence management system docs with architecture diagrams	Enterprise-grade documentation pattern for complex Skills
5	CONTRIBUTING.md	narenkatakam	🟡 Open	Community health improvement (repo at 25% on GitHub metrics)	Infrastructure PR addressing maintainer bandwidth
6	SAP-RPT-1-OSS predictor	amitlals	🟡 Open	Integration with SAP's open-source tabular foundation model for predictive analytics	Bridges enterprise ERP data with Claude Code
7	codebase-inventory-audit	p19dixon	🟡 Open	10-step workflow for identifying orphaned code, unused files, documentation gaps	Technical debt management at scale
8	foundation/standards series (8 skills)	Eaprime1	🟡 Open	tutorial-builder, quiz-generator, a11y-auditor, changelog-curator, data-storytelling, etc.	Broad educational/enterprise enablement push

---

2. Community Demand Trends (From Issues Analysis)

Trend Category	Evidence	Specific Requests
Memory & Persistence	#154, #522, #521, #629	Session survival, context compaction resilience, team knowledge sharing
Security & Governance	#492, #412	Trust boundary protection, agent governance patterns, audit trails
Enterprise Integration	#29, #181	AWS Bedrock compatibility, SAP/ERP connectors, SSO/enterprise auth
MCP Interoperability	#16, #369	Skills-as-MCPs, MCP Apps support in mcp-builder
Skill Quality Tooling	#202, #556	Better skill-creator validation, eval tooling that actually triggers skills
Workspace/Productivity	#299	Google Workspace integration (email, calendar, tasks)

---

3. High-Potential Pending Skills

Skill	PR	Why It May Land Soon
plan-task	#522	Solves #1 user pain (session amnesia); clean implementation with Git-tracked/shared modes
record-knowledge	#521	Companion to plan-task from same author; tagged Markdown persistence
session-memory	#629	Zero-dependency approach to context compaction survival; recent submission with active iteration
ODT skill	#486	ISO standard document format; fills LibreOffice/OpenOffice gap in document skills
management-consulting	#384	Enterprise market expansion; structured frameworks (MECE, Porter's, etc.)
masonry-generate-image-and-videos	#335	Multimodal generation via Imagen 3.0/Veo 3.1; media production workflow

---

4. Skills Ecosystem Insight

The community's most concentrated demand is for stateful agent operation: solving Claude Code's fundamental session amnesia through persistent memory, task continuity, and cross-session knowledge retention—while simultaneously hardening trust boundaries as Skills gain access to sensitive enterprise systems.

---

---

Claude Code Community Digest — 2026-03-25

Today's Highlights

The community is actively addressing critical platform gaps: ARM support for Cowork dominates discussion with 70 comments and 92 upvotes, while a surge of plugin PRs targets long-standing pain points including terminal scroll regressions, memory leaks, and Windows BSODs from parallel filesystem operations. No new official releases shipped in the last 24 hours.

---

Releases

No new releases in the last 24 hours.

---

Hot Issues

Issue	Why It Matters	Community Reaction
#30864: Cowork for ARM processor	Blocks Apple Silicon and ARM Linux users from Cowork's containerized development environment—critical for modern hardware adoption	70 comments, 92 👍; users sharing workarounds, no official timeline
#22016: "Fork conversation from here" broken in VSCode	Core IDE workflow broken for 2+ months; affects conversation management	33 comments, 34 👍; closed today—fix likely shipped
#30640: Native installer fails on FreeBSD	Enterprise/Unix users blocked; bot auto-closed without discussion, reopened by community pressure	32 comments, 52 👍; frustration with support automation
#27897: Cowork VM broken on Windows 11 Insider	EXDEV rename bug persists across updates; affects MSIX installs	21 comments, 9 👍; related to Dropbox/Cloud Files driver conflicts
#3426: MCP tools fail to expose for local Playwright server	Breaks local MCP development workflows; has reproduction	19 comments, 10 👍; active debugging by community
#9177: User-configurable keyboard shortcuts	Long-standing TUX gap; power users need vim/emacs-style customization	14 comments, 26 👍; chronic feature request
#36582: Terminal auto-scrolls to top on long conversations	Severe UX regression making tool unusable for extended sessions	13 comments, 56 👍; community PR submitted (#35683)
#2142: Claude ignores CLAUDE.md security guidelines, exposes API keys	Active security risk; credential leakage to version control	12 comments, 10 👍; reproducible, no fix confirmed
#30854: "Your account does not have access to Claude Code"	Auth regression blocking paid users; unclear error messaging	11 comments, 1 👍; support escalation pattern
#27263: Configurable external URL whitelist for OAuth	Blocks enterprise OAuth flows in App Preview; security/UX tradeoff	11 comments, 45 👍; enterprise-critical

---

Key PR Progress

PR	Feature / Fix	Status
#36614: git-branch-info plugin	Persistent git context display (branch, dirty state, ahead/behind) via SessionStart/UserPromptSubmit hooks	Open
#38182: session-rename plugin with /rename command	Custom session titles via .meta.json sidecars; non-breaking, closes #38173	Open
#38045: lean-agents plugin	Lightweight subagent variants to bypass "prompt too long" from inherited MCP schemas	Open
#35683: scroll-fix plugin	Fixes terminal scroll-to-top regression (#36582) with cursor clamping + freeze toggle	Open
#35710: tool-mutex plugin	Critical: Prevents Windows BSOD (Wof.sys) from parallel filesystem enumeration; fixes #32870	Open
#37236: tmp-cwd-cleanup plugin	Stops /tmp/claude-*-cwd memory leak from Bash tool; closed	Closed
#38105: WhatsApp channel plugin	Bidirectional WhatsApp integration via Baileys (WhatsApp Web protocol); matches Telegram/Discord parity	Open
#27140: memory-bridge plugin	Structured context consolidation at session boundaries; /bridge command for persistent memory	Open
#37159: Persistent memory and session management system	Comprehensive agent memory with context windowing and automatic learning extraction	Open
#31701: Respect $TMPDIR and os.tmpdir()	Fixes hardcoded /tmp/claude paths breaking Termux/Android; closed	Closed

---

Feature Request Trends

1. Cowork Platform Expansion — ARM support (#30864), FreeBSD compatibility (#30640), and relocatable project folders (#38469) dominate; users need Cowork to work across heterogeneous infrastructure.

2. Memory & Context Management — Persistent thinking effort levels (#34171), MEMORY.md line limit increases (#38452), and automated context bridging (#27140, #37159) reflect 1M token window demand.

3. MCP Ecosystem Maturity — Tool exposure reliability (#3426), external URL whitelisting (#27263), and lean agent variants (#38045) show MCP scaling pains.

4. Cross-Device Session Continuity — Mobile access to CLI sessions (#33095), remote Cowork access (#38470), and session renaming (#38182) indicate distributed workflow needs.

5. Voice & Internationalization — /voice language configuration (#31724) for non-English users gaining traction.

---

Developer Pain Points

Category	Issue	Frequency
Windows Stability	BSOD from parallel fs ops (#32870/#35710), Cowork EXDEV rename failures (#27897, #38457), VSCode extension bugs (#22016)	Chronic
Security Gaps	CLAUDE.md guideline adherence failures (#2142, #35199), credential exposure in output (#34819), permission gate bypasses (#33126, #38473)	High
Terminal UX Regressions	Auto-scroll-to-top (#36582), keyboard shortcut inflexibility (#9177)	Acute
Installation/Auth	Platform support gaps (FreeBSD #30640, ARM #30864), cryptic auth errors (#30854)	Persistent
Resource Management	/tmp pollution (#37236, #31701), session loss on directory moves (#33636), remote control TTL issues (#32982)	Recurring

The community is compensating with a plugin ecosystem surge—10+ PRs in 24h targeting core gaps that official releases haven't addressed.

OpenAI Codex — openai/codex

OpenAI Codex Community Digest — 2026-03-25

---

1. Today's Highlights

The Codex team is iterating rapidly on the Rust core with four new alpha releases (v0.117.0-alpha.11–14) in 24 hours, suggesting active stabilization work ahead of a broader release. The TUI is gaining sophisticated new interaction patterns with /btw ephemeral side-threads and /loop scheduling commands, while the community continues to surface critical Windows sandboxing bugs and persistent rate-limit concerns that are generating significant engagement.

---

2. Releases

Version	Notes
rust-v0.117.0-alpha.14	Rapid-fire alpha iteration continues; no detailed changelogs provided—typical for pre-release channels.
rust-v0.117.0-alpha.13	—
rust-v0.117.0-alpha.12	—
rust-v0.117.0-alpha.11	—

Note: No substantive release notes accompany these alphas; developers should expect breaking changes and monitor the repo for migration guidance.

---

3. Hot Issues

#	Title	Why It Matters	Community Reaction
#14593	Still burning tokens very fast with today's VS Code extension update	231 comments, 83 👍 — Business-tier users report aggressive token consumption despite recent patches, suggesting rate-limit accounting remains fundamentally broken for extension users.	High frustration; users sharing detailed usage logs; calls for urgent rollback or compensation.
#10450	Remote Development in Codex Desktop App	439 👍, 78 comments — Top-voted open feature request; gap between VS Code remote workflows and Codex Desktop is blocking adoption for server-side development.	Sustained advocacy; users outlining SSH/container use cases; no official timeline provided.
#11981	Codex app 100% CPU Usage even when only one agent is running	Mac users experiencing thermal throttling and battery drain; indicates background process management issues in Electron/Node runtime.	Repro confirmations across Apple Silicon; temporary workarounds involving process killing.
#13784	Error running remote compact task	"Remote compact task" failures blocking context window management; users losing work when compaction fails mid-session.	Cross-platform reports (Windows, Linux); linked to #14860.
#14860	BUG FIX FOR: "Error running remote compact task: timeout waiting for child process to exit"	Community-submitted fix for Linux; highlights sandbox process lifecycle fragility.	Positive reception for proposed patch; awaiting merge.
#15277	apply_patch fails on Windows after first write in sandbox-owned directory	Regression in VS Code Extension 26.318.11754; Windows developers blocked from iterative file edits.	Critical path bug; no workaround confirmed; affecting daily workflows.
#15503	CPU spikes during normal use	Ubuntu users reporting 100ms-interval debug logging causing measurable overhead; performance regression in latest extension.	Debug logs leaked to production; immediate fix expected.
#15620	High CPU usage when idle: background git/subprocess churn	Root-cause analysis identifies Electron/Node subprocess polling; distinct from #11981 but same symptom class.	Technical deep-dive appreciated; awaiting architectural fix.
#15169	Playwright MCP no longer offers or remembers "approve for this session"	Regression in macOS app build 1088; MCP tool approval UX degraded, forcing repetitive manual confirmations.	7 👍; workflow disruption for browser automation users.
#15680	Codex agent is able to modify its project-level .codex/config.toml without permission	Security concern: sandbox escape allowing self-modifying configuration; undermines trust boundaries.	Zero 👍 but high severity; requires immediate triage.

---

4. Key PR Progress

#	Title	Description	Status
#15701	tui: add ephemeral /btw side-question threads	Introduces /btw command for isolated side-questions without polluting main agent history; includes tui_app_server parity.	Open, fresh (Mar 25)
#15699	[codex] Defer fork context injection until first turn	Optimizes fork startup by delaying build_initial_context until actually needed; reduces memory pressure.	Open
#15597	feat: local ephemeral screen recording	First v1 implementation of local screen capture in codex-app-server; privacy-preserving (process-scoped, ephemeral).	Open
#15531	[hooks] add non-streaming shell-only PostToolUse support	Completes hook system with PostToolUse for shell commands; enables audit logging and reactive controls.	Open, chained behind #15211
#15548	Extract rollout into its own crate	Major refactoring: isolates rollout recorder, session indexing, and policy into codex-rollout; improves build times and modularity.	Open
#15667	[codex] Handle guardian review timeouts explicitly	Increases guardian timeout 90s→120s; adds TimedOut terminal status with proper TUI surfacing (not silent rejection).	Open
#15693	fix: fix old system bubblewrap compatibility	Resolves Linux sandbox failures on older bubblewrap builds (rejects --argv0); prefers system bwrap, falls back to vendored only when missing.	Open, fixes #15283
#15695	Migrate config loading to be async	Foundational async refactor starting with resolve_root_git_project_for_trust; unblocks future I/O-bound configuration features.	Open
#15601	[app-server] Add a method to override feature flags	Enables global (not just thread-level) feature flag overrides; critical for staged rollouts and enterprise policy.	Open
#15691	feat: exec-server prep for unified exec	Architectural preparation for merging execution paths; reduces divergence between CLI and app-server runtimes.	Open

---

5. Feature Request Trends

Theme	Evidence	Momentum
Remote/SSH Development	#10450 (439 👍)	Highest-voted open request; enterprise blocker
Scheduling & Automation	#8317 (time-based commands), #15679 (/loop command), #14110 (persistent agent)	Emerging pattern: users want Codex as daemon, not just interactive tool
Hook/Extension Points	#14754 (PreToolUse/PostToolUse), #15531	Strong demand for code quality gates and compliance automation
Transparency & Changelogs	#4323 (66 👍)	Longstanding documentation gap for IDE extension
Keyboard-First TUI	#14939 (YOLO mode shortcut), #15679	Power users seeking vim-like efficiency

---

6. Developer Pain Points

Issue	Frequency	Impact	Current Status
Rate limit transparency & burn rate	Daily reports	🔴 Critical	#14593 active; no resolution
Windows sandbox/tool-call reliability	Weekly regressions	🔴 Critical	#15277, #13905, #10187; patchy fixes
Idle CPU/battery drain (macOS/Linux)	Multiple reports	🟡 High	#11981, #15620; root causes identified, fixes pending
Context compaction failures	Sporadic but severe	🟡 High	#13784, #14860; community fix proposed
MCP/tool approval UX friction	Regressions noted	🟡 Moderate	#15169; session persistence broken
Configuration escape hatches	Security concern	🔴 Critical	#15680; newly reported, unassigned

---

Digest compiled from github.com/openai/codex public activity. For real-time updates, subscribe to release notifications or join GitHub Discussions.

Gemini CLI — google-gemini/gemini-cli

Gemini CLI Community Digest — 2026-03-25

---

1. Today's Highlights

The v0.36.0-preview.0 release introduces multi-registry architecture and tool filtering for subagents, enabling more sophisticated agent orchestration workflows. The community is actively debating workspace boundary security after a new bug report highlighted inconsistent user-approval workflows between bash and declarative tools. Meanwhile, Google Summer of Code 2026 preparations are underway with a high-engagement tracking issue for behavioral evals and OSS community improvements.

---

2. Releases

v0.36.0-preview.0

• Multi-registry architecture and tool filtering for subagents — enables subagents to operate with constrained toolsets across different registries (PR #22712)

v0.35.0

• Customizable keyboard shortcuts — users can now remap key bindings to their preferences (PR #21945)
• AgentLoopContext threading — core infrastructure improvement for agent state management (PR #21944)

---

3. Hot Issues

Issue	Why It Matters	Community Reaction
#23331 GSoC: Behavioral evals, Quality, and OSS Community	52 comments — central coordination for 2026 GSoC participation, focusing on making quality contributions accessible to external developers	High engagement; 14 👍; seen as critical for scaling community contributions
#23328 Optimize test suite output logging	41 comments — test verbosity creates hundreds of thousands of characters of noise, slowing CI feedback loops	9 👍; developers frustrated by debugging friction
#23717 Bug: Inconsistent User-Approval Workflows for Workspace Boundaries	Security-critical: bash tool prompts for out-of-workspace access, but declarative tools bypass approval — creates trust boundary violations	Fresh report (2 comments); likely to escalate
#23230 Exiting plan mode does not switch model	Plan-to-execution model downgrade (Gemini 3 Flash) fails silently, wasting expensive model calls	10 comments; affects cost-sensitive workflows
#21109 Animated waveform visualizer for voice mode	GSoC Project 11 deliverable; voice mode lacks state feedback, hurting UX	9 comments; part of larger voice mode maturity push
#22745 Assess AST-aware file reads, search, and mapping	EPIC: Could reduce token waste from misaligned file reads and improve codebase navigation precision	4 comments; technical evaluation in progress
#23728 Steer sub-agents	Model steering (temperature/persona) currently unavailable for subagents, limiting specialized agent behaviors	Just opened; fills critical orchestration gap
#23724 Persistent Project-Level Tracker Storage	Moves task state from ephemeral /tmp to .gemini/tracker/, enabling Git-trackable collaboration	Part of SDD Phase 3; DAG-based planning replacing Markdown checklists
#22855 Support passing prompt to /plan	/plan currently requires interactive mode; single-command planning would streamline automation	2 comments; workflow efficiency request
#23582 Subagents Awareness of Active Approval Modes	Subagents lack context on Plan vs. Auto-Edit mode, causing instruction/policy conflicts	1 comment; architectural consistency issue

---

4. Key PR Progress

PR	Feature/Fix	Impact
#23732 + #23731	Move flaky Ctrl+C test to non-blocking suite	CI stability: Unblocks PRs from macOS-specific signal handling flakes
#23729	Conditionally expose additional_permissions in shell tool	Security: Hides sandboxing complexity from model when disabled; removes hardcoded policy check
#23719	Safe and rich themes with true color support	UX: Fixes 256-color consistency issues; adds high-visibility mappings for accessibility
#23720	CI Skill for automated failure replication	Developer velocity: Enables local reproduction of CI failures without passive waiting
#23727	Behavioral eval for web tool selection	Quality: Tests web_fetch vs. google_web_search discrimination — closes eval coverage gap
#23282	forbiddenPaths for OS-specific sandbox managers	Security: Cross-platform sandbox hardening (macOS/Linux/Windows) with unified test coverage
#23725	Increase nightly eval runs 3→10	Reliability engineering: Eliminates "noisy floor" for statistical confidence in quality signals
#22869	Dynamic toggle for alternate buffer mode	UX: Switch inline/fullscreen mid-session without restart
#23064	Redesign context and compression UI	UX: Less intrusive auto-compression; subtle gray italic styling with percentage indicators
#23691	Windows sandbox dynamic expansion Phase 1/2.1	Platform parity: Network access and Low Integrity write permissions for Windows sandbox

---

5. Feature Request Trends

Direction	Evidence	Momentum
Agent orchestration & subagent control	Multi-registry architecture, steering requests, approval mode awareness, memory routing	🔥 High — core architectural evolution
Persistent, collaborative state	Project-level tracker storage, memory routing (global vs. project), SDD DAG tasks	🔥 High — shifting from session-based to project-native
Voice mode maturity	Waveform visualizer, Live API PoC, GSoC project	🟡 Medium — active but scoped
AST-aware tooling	File read bounds, codebase mapping, method-level precision	🟡 Medium — investigation phase
Theme/accessibility polish	True color support, safe 256-color themes, dynamic buffer toggle	🟡 Medium — quality-of-life focus
Security boundary consistency	Workspace approval workflows, sandbox expansion, forbiddenPaths	🔥 High — trust-critical, actively patched

---

6. Developer Pain Points

Pain Point	Frequency	Status
Test suite noise & CI flakiness	Recurring (#23328, #23731-23732, #23313)	Active mitigation; tests being moved to non-blocking
Model/context compression opacity	Recurring (#23556 pathological compression, #23064 UI redesign)	UI improvements shipped; root causes under investigation
Subagent limitations	High (#23728 steering, #23582 approval awareness, #22745 AST tools)	Architectural work in progress
Workspace security inconsistency	Emerging (#23717)	Bug report filed; likely priority escalation
Plan mode workflow friction	Steady (#23230 model switch, #22855 inline prompts, #23577 terminal height)	Iterative improvements; no holistic fix yet
Extension/startup noise	Steady (#23175 duplicate warnings)	PRs in flight

---

Digest compiled from google-gemini/gemini-cli public activity. For real-time updates: github.com/google-gemini/gemini-cli

GitHub Copilot CLI — github/copilot-cli

GitHub Copilot CLI Community Digest — 2026-03-25

1. Today's Highlights

No new releases dropped in the last 24h, but the community is actively wrestling with MCP server policy enforcement bugs and model availability discrepancies between CLI and VS Code. A surge of fresh issues around terminal compatibility (Windows Terminal scrollbars, Warp support) and session management suggests users are pushing Copilot CLI into more diverse workflows.

---

2. Releases

No releases in the last 24 hours.

---

3. Hot Issues

#	Issue	Why It Matters	Community Reaction
#1703	Copilot CLI does not list all org-enabled models (e.g. Gemini 3.1 Pro)	Critical parity gap: Same org, same account, different model lists between CLI and VS Code. Blocks enterprise users who rely on latest models.	22 👍, 13 comments — highest engagement; users confirming org settings are correct
#501	~/.bash_history truncated	Long-running bug where CLI spawns shells without loading .bashrc, destroying user history configurations. Finally closed.	10 comments, mixed relief; fix validation ongoing
#172	MCP timeouts not respected	Breaks long-running MCP servers; mcp-config.json timeout field ignored. Blocking for custom tooling workflows.	7 comments, 2 👍; workaround requests
#2236	Org MCP servers disappear — false "disabled by organization" warning	Regression in v1.0.11: Servers worked last week, now falsely blocked. Org policy drift between CLI and VS Code.	7 👍, 3 comments; urgent for enterprise
#2050	Claude Sonnet 4.6 — HTTP/2 GOAWAY / 503 errors	Model-specific reliability issue; retry logic failing after 84s. Gemini 3 Pro works fine — points to CAPI routing.	6 comments, 3 👍; reproducible reports
#1477	"Continuing autonomously" charges premium requests after completion	Billing/UX concern: Users charged for "autonomous" continuation after model already finished. Feels like double-dipping.	7 👍, 6 comments; frustration with pricing transparency
#989	--acp mode returns incorrect tool IDs	Breaks ACP protocol compliance; tool call IDs use titles instead of UUIDs. Affects client integrations.	5 comments, 3 👍; Microsoft employee reported
#1257	Termux/Android support	Mobile developer demand; pty.node native module fails on Android. Expands addressable user base significantly.	6 👍, 4 comments; workaround attempts shared
#2208	WSL CPU spin + unresponsive TUI	Performance regression with large state (Claude Opus 4.6 1M context). Trace collected.	1 comment; needs investigation
#2252	Windows Terminal scrollbar missing in v1.0.11	Fresh regression: Breaks basic terminal navigation. Ctrl+Shift+Up also broken.	New issue, immediate reports

---

4. Key PR Progress

No pull requests updated in the last 24 hours.

---

5. Feature Request Trends

Based on open issues, three clear demand vectors emerge:

Trend	Evidence	User Need
Granular permission controls	#2256, #2273, #899	Whitelist binaries (ls, grep), distinguish read-only vs. destructive PowerShell commands, file-type filters
Session persistence & automation	#2227, #2268, #2108	Auto-share on exit, launch flags for plan/autopilot mode, session_id in hooks for external tooling
Cross-platform terminal parity	#1257, #2246, #2271	Android/Termux, Warp terminal, configurable shell backend (bash vs. PowerShell on Windows)

---

6. Developer Pain Points

MCP Policy Drift — The most acute frustration: CLI and VS Code interpret organization Copilot policies differently. Users report #1707, #2236 servers working in VS Code but blocked in CLI, or disappearing after version bumps. Suggests policy caching or versioned enforcement logic needs alignment.

Model Availability Inconsistency — #1703 highlights that CLI lags VS Code in exposing org-enabled models (Gemini 3.1 Pro). Creates confusion about whether issues are policy, licensing, or client-side.

Permission Fatigue — Repeated approvals for ls, grep, read-only PowerShell commands #2256, #2273 disrupt flow. Users want persistent whitelists or read-only command auto-approval.

Performance at Scale — #2208, #2269 report CPU saturation with large contexts (1M tokens) or multiple SDK sessions. 4-core Codespaces become unusable — resource efficiency gap for CI/automation use cases.

Line Ending Corruption — #1148 — CLI forces CRLF on LF files, polluting repositories. Cross-platform development friction.

Kimi Code CLI — MoonshotAI/kimi-cli

Kimi Code CLI Community Digest — 2026-03-25

---

1. Today's Highlights

The v1.25.0 release is driving significant community activity, with multiple bug reports surfacing around the writefile tool and the new plugin system prompting both questions and documentation updates. Meanwhile, two critical PRs merged to resolve hanging issues in yolo/print mode and restore Jinja2 include support for system prompts, improving reliability for automation workflows.

---

2. Releases

No new releases in the last 24 hours.

---

3. Hot Issues

#	Issue	Status	Why It Matters
#1564	writefile tool frequently errors after v1.25.0 upgrade	OPEN	Core workflow regression affecting Windows users; partial workaround identified (chunked writes) but root cause unaddressed
#1565	WriteFile/Shell fail with "Invalid arguments" on HTML with quotes	OPEN	Blocks basic web development workflows; argument parsing appears to mishandle quoted strings
#1334	JetBrains IDE integration broken	OPEN	Long-running IDE compatibility issue; affects PyCharm/IntelliJ users with no resolution timeline
#1566	Plugin system compatibility questions (Claude Code/Open Code plugins)	OPEN	Community seeking clarity on plugin ecosystem interoperability; documentation gap evident
#1567	Disable Ctrl+W clipboard behavior on Mac	OPEN	UX friction: destructive clipboard override conflicts with standard terminal expectations
#1560	VS Code plugin: Skill selection should not auto-submit	OPEN	Inconsistency between CLI and VS Code behavior; breaks iterative skill refinement workflow
#1559	Official download command fails	OPEN	Onboarding friction: getting-started documentation contains broken installation instructions
#1380	ACP terminal tool fails with module acp has no attribute TerminalHandle	CLOSED	ACP 0.8 API breaking change; resolved via #1400 migration
#1476	Plan mode state desync: "no longer in plan mode" while still in plan mode	CLOSED	State management bug; fix likely shipped in recent release

---

4. Key PR Progress

#	PR	Status	Feature/Fix
#1562	Auto-resolve QuestionRequest in yolo/print mode	MERGED	Eliminates indefinite hangs in automation pipelines; adds YoloModeInjectionProvider to suppress interactive prompts
#1563	Jinja2 {% include %} support for system prompts	MERGED	Enables modular prompt composition; critical for teams managing complex agent configurations
#1561	Lifecycle hooks system (Wire 1.7)	OPEN	13-event hook architecture matching Claude Code; enables custom telemetry, logging, and automation integrations
#1549	Configurable compaction providers	OPEN	Allows dedicated compaction models vs. chat model reuse; cost/performance optimization for long contexts
#1512	Fix ACP init when kimi not in sys.argv	OPEN	Resolves IDE integration failures (PyCharm, VS Code debugger); addresses list.index(x): x not in list error
#1400	Replace deprecated acp.TerminalHandle with Client methods	MERGED	ACP 0.8 compatibility; unblocks terminal tool functionality
#1558	Mark plugin system as beta	MERGED	Documentation transparency; sets expectations for API stability
#1213	Nix devShell support	OPEN	Reproducible development environment for Nix users; community infrastructure contribution

---

5. Feature Request Trends

Trend	Evidence	Implication
Plugin ecosystem clarity	#1566, #1558	Users want interoperability standards and installation guidance for third-party plugins
Lifecycle extensibility	#1561	Demand for hooks/observability matching competitor (Claude Code) capabilities
Context management optimization	#1549	Need for cost-efficient compaction strategies in long-running sessions
Cross-IDE parity	#1334, #1560, #1512	VS Code and JetBrains integrations require dedicated attention to match CLI reliability

---

6. Developer Pain Points

Pain Point	Frequency	Impact
v1.25.0 writefile regressions	2+ reports (#1564, #1565)	Blocking daily coding workflows; Windows and HTML content particularly affected
Argument parsing fragility	#1565	Tool calls failing on valid content with special characters
IDE integration gaps	#1334, #1512, #1560	JetBrains broken, VS Code behavior inconsistent, debugger contexts fail ACP init
Documentation drift	#1559, #1566	Installation instructions outdated; plugin system under-documented
Mac UX friction	#1567	Unexpected clipboard behavior violating platform conventions

---

Digest compiled from github.com/MoonshotAI/kimi-cli activity for 2026-03-25.

OpenCode — anomalyco/opencode

OpenCode Community Digest — 2026-03-25

---

1. Today's Highlights

OpenCode shipped v1.3.2 with critical TUI debugging capabilities including heap snapshot functionality, while the community converged on memory leak fixes following a coordinated 20+ contributor investigation. Authentication turbulence continues with Claude OAuth breakage affecting all models since March 17, and v1.3.2's web interface suffered a partial outage even for users on older versions—suggesting a server-side regression.

---

2. Releases

v1.3.2

• TUI debugging: Added heap snapshot capture for both TUI and server processes via "Write heap snapshot" command; outputs to tui.heapsnapshot and server.heapsnapshot

v1.3.1

• Poe integration: Built-in OAuth and API key authentication for poe.com
• AWS improvements: Token caching for custom Amazon Bedrock providers
• Editor support: Syntax highlighting for Kotlin, HCL, Lua, and TOML
• UX polish: Manual lock/unlock for theme mode

---

3. Hot Issues

Issue	Why It Matters	Community Reaction
#16470 — Code output unreadable in light mode	Core accessibility bug making formatted code invisible; affects daily usability	24 comments, 26 👍 — top-voted active bug
#16697 — Multiple memory leaks, unbounded RAM growth	Critical stability issue; 20+ contributors independently investigated, resulting in coordinated PR wave	14 comments, 11 👍; flagged as systemic problem
#17910 — OAuth auth + cache_control ephemeral causes HTTP 400 on Claude models	Complete Claude outage for OAuth users since March 17; regression with no config changes	11 comments; urgent, affects paying subscribers
#4804 — High CPU usage (closed)	Long-standing performance issue finally resolved; Mac Intel users particularly affected	17 comments, resolved after months
#19034 — opencode web completely broken since 1.3.2	Severe: Users on 1.2.27 also affected, implying server-side/API regression; models won't load, sessions lost	6 comments, rapid closure suggests hotfix deployed
#7467 — GitHub-based Agent Marketplace	Ecosystem growth enabler; addresses friction in sharing agents across teams	10 comments; sustained interest since January
#11176 — Official VS Code extension	Most-requested platform expansion (23 👍); would unlock native IDE integration	9 comments, strong demand
#17418 — Remote server paths broken on Windows since 1.2.25	Breaks cross-platform remote workflows; Windows Desktop → Linux server path confusion	8 comments; regression in recent release
#18950 — Cannot connect with Claude (web UI)	Auth flow regression: OAuth options disappeared, leaving only API key input	4 comments; UX confusion post-update
#18364 — Windows TUI crashes during input, editing state lost	Data loss bug: crashes destroy unsaved work; high severity for Windows users	4 comments

---

4. Key PR Progress

PR	Description	Status
#18679 — Effectify Worktree service	Major architecture migration: Worktree service to Effect pattern with ChildProcessSpawner for git ops	Open, beta, vouched
#18900 — Restore git-backed review modes with Effectful git service	Restores desktop diff inspection; unifies git operations across VCS, worktree, CLI	Open
#19041 — Auto-re-read stale files in agent loop	Fixes retry loops when formatters modify files between read/write; prevents hundreds of consecutive failures	Open, needs compliance
#18419 — Switch file search to fff	Replaces ripgrep plumbing with unified fff layer for search/tree/grep; performance and consistency improvement	Open
#6629 — OpenTelemetry instrumentation with Aspire Dashboard	Full observability: tools, MCP, sessions, LLM, LSP, plugins; bun run dev:otel for local debugging	Open, needs issue
#19038 — In-app browser for desktop	Native web browsing without leaving OpenCode Desktop	Open
#18433 — AI SDK v6 support	Major dependency upgrade; WIP with copilot and GitLab AI SDK migrations	Open, beta, vouched
#16069 — First-class PowerShell support on Windows	Replaces Git Bash realpath dependency; native PowerShell path/permission handling	Open
#18767 — Mobile touch optimization	Responsive touch targets, swipe gestures, mobile-friendly navigation while preserving desktop UX	Open
#19035 — Copy format options (Markdown, Email-ready HTML)	Replaces single copy button with dropdown: plain text, Markdown, HTML; improves sharing workflows	Open

---

5. Feature Request Trends

Trend	Evidence	Implication
IDE ecosystem expansion	#11176 (VS Code ext, 23 👍), #7467 (Agent Marketplace)	Users want OpenCode inside existing workflows, not as siloed replacement
Remote/mobile access	#19013 (WebRTC P2P remote), #18767 (mobile touch)	Demand for "code anywhere" flexibility beyond desktop tethering
Conversation/memory tools	#11819 (search message history), #15351 (chat channels)	Users hitting context limits; need better memory/organization primitives
TUI polish & accessibility	#19005 (clickable paths), #18195 (keyboard scroll), theme fixes	Terminal UX still rough vs. web; friction in daily workflows
Voice & output quality	#19025 (Edge TTS server backend)	Browser TTS insufficient; users want production-grade voice output

---

6. Developer Pain Points

Pain Point	Frequency	Root Cause	Community Sentiment
Authentication fragility	Very high	Claude OAuth breakage (#17910), Poe/Claude API key confusion (#18950, #18811)	Frustration with "moving target" auth flows; enterprise users particularly affected
Memory & stability	High	Multiple leak sources (#16697), unbounded RAM growth, compaction loops (#18962)	Critical mass reached: 20+ contributors independently investigating; PR wave emerging
Theme/rendering regressions	High	Light mode breakage (#16470), system theme transparency issues (#12184, #13363)	Accessibility debt; rapid TUI iteration causing visual regressions
Cross-platform path handling	Moderate	Windows path separators (#7279), remote server path confusion (#17418)	Windows and remote workflows second-class; PowerShell PR (#16069) addresses part
MCP server management	Moderate	No runtime toggle (#514), startup errors with unavailable servers	Context bloat and reliability issues; needs TUI controls
Version synchronization	Moderate	v1.3.2 web breakage affecting 1.2.27 users (#19034)	Suggests insufficient server/API versioning; breaks trust in "stable" releases

---

Qwen Code — QwenLM/qwen-code

Qwen Code Community Digest — 2026-03-25

---

1. Today's Highlights

The Qwen Code team shipped v0.13.0-nightly.20260324 with continued iteration on the ACP (Agent Client Protocol) integration and permission system refinements. Community activity surged around VS Code extension UX gaps—particularly clipboard image pasting and settings discoverability—while developers continue hitting edge cases with CJK (Chinese/Japanese/Korean) path handling and shell environment detection on Windows/Linux. Notably, a new /loop command for periodic task execution and Telegram channel integration via ACP signal expansion into multi-platform agent deployment.

---

2. Releases

Version	Date	Notes
v0.13.0-nightly.20260324.aebe889b3	2026-03-24	Nightly build; full changelog available here. No detailed release notes provided—typical for nightly channel.

---

3. Hot Issues

#	Issue	Why It Matters	Community Reaction
#1370	VSCode extension settings discovery gap	Critical onboarding friction: users cannot locate settings UI or confirm configuration parity between CLI and IDE extension.	6 comments, sustained activity since Dec 2025; indicates documentation/UX debt.
#241	Shift+Enter sends message instead of newline	Breaks muscle memory from Claude Code and standard chat interfaces; impacts power user efficiency.	4 comments, 4 upvotes—long-standing (Aug 2025) with clear user expectation mismatch.
#2626	CLI crashes on permission prompts (Linux/tmux)	Regression affecting Ubuntu Server 24.04.4 users; blocks core workflow in headless environments.	4 comments, zero workarounds identified beyond version rollback.
#2632	Responses hang indefinitely (10–15 min)	Severe reliability issue requiring manual ESC interruption; affects trust in agent autonomy.	3 comments, Chinese-language report suggests regional user base impact.
#1897	LLM injects spaces into CJK paths	Recurring localization bug breaking file operations in Chinese-named directories; path validation fails.	3 comments, duplicate pattern with #2496 indicating systemic tokenizer/prompt issue.
#2605	Windows clipboard image paste unsupported	Feature parity gap vs. macOS/Linux; Windows CLI users forced to manual path entry.	4 comments, clear feature request with workflow justification.
#2640	Permission approvals not persisting in session	"Always allow" selections ignored within same session; breaks YOLO mode promise and creates fatigue.	5 upvotes, 1 comment—high impact, likely regression from permission system refactor.
#2530	"Empty response text" errors with OpenRouter	Integration reliability with third-party providers (Qwen3-Coder-next, GLM-5-turbo); tool calling failures.	2 comments, affects multi-model workflows.
#2629	VS Code: file_path required property missing	Recent update broke file read operations; params must have required property 'file_path' regression.	1 comment, immediate post-release report suggests 0.13.0 instability.
#2304	<think> tag causes VS Code extension truncation	Thinking/reasoning content malformed in streaming; wastes API calls and breaks output.	1 comment, Chin

---

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