Radar brief — 2026-08-22

What the evidence shows

Versioned GitHub releases dominate this window (Codex, Claude Code, Goose, Cline, OpenCode, Antigravity, Gemini nightlies). Themes cluster on CLI/agent orchestration, model/provider routing, MCP/skills surfaces, and cost/usage signals—with several breaking / prerelease tags.

Low-confidence / noisy: Gemini entries are nightlies (fast-moving, may not ship as-is). Cline desktop handoff is beta + feature-flagged. Windsurf signal is changelog-HTML (score 147 vs 190–250 GitHub releases)—treat as directional, not pin-equivalent.

For developers

• On Codex CLI: Upgrade through rust-v0.148.0 → rust-v0.149.0 if you live in the TUI. Practice /export, fork/resume, and the new codex agents + codex queue flow before relying on them mid-incident. Turn on cost in /status where eligible so thread spend is visible without leaving the terminal.
• Bedrock users: rust-v0.148.0’s built-in Bedrock Runtime path is worth a dry run (profile, region, model routing) in a throwaway workspace before swapping defaults.
• Goose v1.46.0: Use per-message usage (tokens/cost/TTFT/tok/s) to baseline slow vs expensive turns; streaming shell output changes how you watch long commands—verify scripts that assumed buffered output.
• Cline SDK v0.0.76: Prefer skill slash commands via the skills tool (cleaner transcripts). Try scheduled tasks/todos only in a single workspace first (schedules are workspace-scoped). Image generation: confirm models actually support it before designing flows around it.
• OpenCode v1.18.19: If you use Cloudflare AI Gateway or ChatGPT-tied Codex limits, re-check /connect, pricing display, and rate-limit behavior after upgrade; malformed pricing no longer should zero out cost math.
• Gemini CLI nightlies: Seatbelt socket isolation and symlink ignore fixes are security/correctness wins—test on a non-prod machine; don’t pin production automations to nightlies.
• Zed v1.15.0: Set git.diff_base explicitly so gutters match your review habit (uncommitted vs branch-vs-default).
• Claude Code v2.1.234: If your host gives each session its own config dir, set CLAUDE_CODE_PROJECT_DIR_NAME so transcript paths stay short and stable; bind selection:clear if you select heavily in the agents view.
• Antigravity 1.1.12: Use t outline on long artifact markdown instead of scroll hunting.

For teams

• Multi-session ops (Codex): Stand up a short pilot of the agents dashboard + queue (see open proposal) for on-call or parallel coding agents—define who may start/stop/rename shared tasks and how remote vs local sessions are named.
• Cost governance: Codex thread credits in status + Goose per-message/session cost UIs are enough to require “cost visible in default status line” in team runbooks; capture before/after on a fixed workload.
• Cline Cloud handoff (beta): Treat /handoff as experimental. Require clean git preflight (repo/branch/commit pushed), document what leaves the laptop (transcript, images, follow-up command), and gate the Cloud feature flag to a volunteer cohort.
• MCP estate: Windsurf’s self-hosted GitLab / older Atlassian MCP auth fixes and Codex/Goose/Cline MCP-tagged surfaces mean re-verify corporate MCP servers after client bumps—especially non-cloud forges.
• Sandbox policy: Track Gemini’s macOS Seatbelt container-socket isolation as a pattern: agents that can see Docker sockets are a lateral-risk class—align desktop agent sandbox rules with that threat model.
• IDE baseline: Zed’s git.diff_base should be a shared editor default so PR review colors mean the same thing across the team.
• Hosting Claude Code: Platforms that isolate per-session config should standardize CLAUDE_CODE_PROJECT_DIR_NAME and GitLab glab auth so MR badges and transcript paths are consistent in shared images.

What to ignore

• Marketing volume (“+160 more changes”)—act only on named user-facing capabilities in the table.
• Pinning Gemini nightlies or Cline desktop beta handoff as production standards.
• Windsurf sidebar perf unless you actually hold thousands of cached sessions; the MCP auth fix is the sharper operational takeaway, with lower evidence weight than GitHub tags.
• Open proposals at P5 noise unless they match your stack; the Codex dashboard/queue proposal is the one aligned with this brief’s strongest release cluster.
• Theme tag inflation (breaking × 8, mcp × 7): tags are classifiers, not proof every release breaks you—read the named bullets.

Watch next

• Codex agents dashboard / queue adoption and any follow-on permission or sandbox constraints after rust-v0.149.0.
• Whether Cline Cloud handoff exits beta and how workspace/schedule scope interacts with multi-root repos.
• Goose unrolled loop + cache-safe turns—stability and cost under long agent sessions.
• Gemini CLI Seatbelt sandbox work graduating from nightly to a stable tag.
• OpenCode gateway passthrough + rate-limit parity under real ChatGPT subscription load.
• Open proposals: Codex multi-session control (team); Grok CLI prompt/MCP cleanups; OpenCode PR #38252 verification videos (P2).

How to use DevAgentRadar

• Read What the evidence shows first—only versioned rows are claims; everything else is judgment on those rows.
• Developers implement one client upgrade + one workflow change (export/queue, usage UI, skills tool, diff_base).
• Teams turn the same signals into policy: cost visibility, MCP re-verify, beta feature flags, sandbox socket rules.
• Ignore source-weak or prerelease pins until they reappear as stable tags.
• Feed outcomes back via improvement proposals when a release changes how your org should run agents—not when a changelog is merely long.

Evidence appendix

• Evidence window: top curated signals (versioned releases preferred)
• Assistants tracked: 22
• Signals in this brief: 12
• Trusted source mix (GitHub/changelog vs HTML): 55%
• Open improvement proposals: 4
• Sources reporting errors: 0

High-signal releases used

• OpenAI Codex CLI rust-v0.148.0: Export complete TUI conversations to Markdown with /export, either to the clipboard or a new file. (#37358) · Fork sessions with codex exec fork, and archiv — source

• Claude Code v2.1.234: Added the optional CLAUDE_CODE_PROJECT_DIR_NAME environment variable: hosts that give each session its own config directory can choose a short name for the pe — source

• OpenAI Codex CLI rust-v0.149.0: Added an interactive codex agents dashboard for searching, starting, opening, renaming, and stopping tasks, with configurable shortcuts. (#39094, #39112, #391 — source

• Zed AI v1.15.0: This week's release includes a new git.diff_base setting for choosing whether editor gutters, file-status colors, and git: diff show uncommitted changes aga — source

• Goose v1.46.0: Unrolled agent loop #9574 · Cache-safe request assembly with append-only turn context and declared cache semant — source

• Cline sdk/sdk/v0.0.76 — SDK v0.0.76: Added model-driven image generation. · Models that support it can generate images during a turn, and generated images are persisted in session history and expor — source

• Cline desktop-v0.0.15-beta.1 — beta.1: Beta: hand off a local session to Cline Cloud with /handoff — the conversation, attached images, and an optional follow-up command move to a cloud workspace t — source

• Google Antigravity 1.1.12: Added a heading outline to the artifact viewer, opened with t, so a long markdown document's structure is visible at a glance and you can jump straight to a s — source

• Gemini CLI v0.56.0-nightly.20260821.g30573d2e4: fix(core): ensure consistent symlink evaluation in ignore path handling · refactor(core): remove eslint-disable and type-asserts from shellExecutionService — source

• Gemini CLI v0.56.0-nightly.20260822.g5411f113c: fix(sandbox): isolate Docker and container runtime sockets and binaries in macOS Seatbelt — source

• OpenCode v1.18.19: Added native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway models. · Matched Codex rate limits more closely to ChatGPT subscription limits. (@Game — source

• Windsurf v3.7.25: v3.7.25 August 13, 2026 Devin Desktop Faster sidebar for users with thousands of cached sessions: filtering, space grouping, and sorting now only process the se — source

Open proposals

• Stand up codex agents dashboard + queue for multi-session task control (audience: team, P5)
• Replace <output_efficiency> with <response_guidelines> in Grok CLI prompts (audience: developer, P5)
• Drop legacy managed MCP configs and shadowed platform skills in Grok CLI (audience: everyone, P5)
• Improve with OpenCode pr-38252-videos: pr-38252-videos PR #38252 verification videos (audience: everyone, P2)