Radar brief — 2026-08-23

What the evidence shows

Versioned GitHub releases dominate this window (Codex, Claude Code, Goose, Cline, OpenCode, Antigravity, Zed). Windsurf is changelog/HTML; Gemini entries are nightlies—treat those as directional, not ship gates.

Theme read: cli / agent / mcp / model cluster around session orchestration, cost telemetry, skills loading, and sandbox/MCP trust boundaries—not greenfield model drops alone. breaking tags are dense on Codex, Cline SDK, Zed, Gemini nightlies, and Windsurf; verify upgrade notes before fleet rollout.

For developers

• Codex (0.148 → 0.149): Practice /export for reviewable thread artifacts; use fork + resume picker instead of re-prompting long contexts. Turn on cost/credits in /status before large refactors. Try codex agents + codex queue when juggling parallel local/remote tasks; learn /cd//pwd so cwd drift does not silently retarget tools.
• Goose 1.46.0: Lean on streaming shell and per-message cost/TTFT stats to catch runaway loops early; cache-safe turn assembly matters if you rely on prompt caching.
• Cline SDK 0.0.76: Prefer skill slash commands (skills tool path) so transcripts stay small; use workspace-scoped schedules/todos for recurring chores; only enable image generation where the model actually supports it and you want binaries in history/exports.
• OpenCode 1.18.20: On subagent failure, resume via surfaced task_id instead of restarting empty-handed; expect permission prompts mid-opencode run when children request tools.
• Claude Code 2.1.234: Set CLAUDE_CODE_PROJECT_DIR_NAME if your host isolates config dirs per session; bind selection:clear; wire glab auth to get MR state in the statusline.
• Zed 1.15.0: Set git.diff_base explicitly so gutter/diff match your branch review habit (uncommitted vs merge-base).
• Antigravity 1.1.12: Use outline key t on long artifact markdown before scrolling.
• Gemini nightlies: Note Seatbelt socket isolation and symlink/ignore fixes—validate local sandbox + ignore behavior before depending on them in daily drivers.
• Windsurf 3.7.25: If MCP to self-hosted GitLab/older Atlassian failed auth, retest after upgrade; sidebar perf fix helps heavy session caches.

For teams

• Multi-session control: Pilot Codex agents dashboard + queue (see open proposal) as a shared pattern for “many tasks, one operator” before standardizing runbooks. Pair with export-to-Markdown for audit-friendly handoffs.
• Cost governance: Goose per-message usage/cost and Codex thread credits are concrete hooks for budget norms—require status/cost visible in CI-adjacent or long agent jobs.
• Skills & MCP hygiene: Cline’s skills-tool loading (no double-paste of SKILL.md) and desktop Plugins/Marketplace consolidation reduce prompt bloat and shadow configs. Windsurf MCP auth fixes matter for GitLab/Atlassian-backed enterprise servers—add to MCP allowlist regression tests.
• Desktop rename (Cline): Communicate the Cline Code → Cline name/icon change; confirm settings/sessions/credentials migration in managed images so support does not treat it as a new install.
• Sandbox/security: Track Gemini Seatbelt isolation of container runtime sockets as a pattern for macOS agent hosts; do not fleet-wide on nightlies until a stable tag lands.
• IDE policy: Zed git.diff_base should be documented in team editor defaults so review diffs mean the same thing across machines.
• Open proposals to staff: Codex agents dashboard + queue (team, P5); Grok CLI prompt/MCP cleanup items (developer/everyone, P5); OpenCode PR verification videos (everyone, P2).

What to ignore

• Marketing-only framing around renames and “Recommended/Free” tier labels until your billing path is verified—Cline’s selector UX change is real, but tier economics are not fully specified in the release text.
• Gemini nightly tags as production pin points; useful security/symlink signals, still prerelease.
• Windsurf HTML changelog relative to GitHub-tagged peers when you need reproducible artifact hashes—use it for product direction, prefer tagged repos for automation.
• “+N more changes” tails without reading the linked release notes—score volume ≠ upgrade urgency.
• Inflated cross-assistant theme counts (model×10, cli×10): they reflect tag density in this curated set, not equal shipping weight.

Watch next

• Codex post-0.149 adoption of agents / queue and whether Bedrock routing + credit telemetry stay accurate under load (proposal: stand up dashboard + queue).
• Goose cache semantics + unrolled loop behavior under long MCP/tool traces.
• Cline schedule/todo durability across machines and whether image outputs stay manageable in shared session stores.
• OpenCode resumable subagent task_id + subagent permission UX under opencode run (related improvement proposal on verification videos).
• Gemini CLI promotion of Seatbelt container-socket isolation from nightly → stable.
• Grok CLI open proposals on prompt guidelines and legacy managed MCP/skills cleanup (not in this evidence set’s release rows, but queued at P5).
• Error-reporting sources: currently 0—recheck if MCP auth or sandbox isolations start failing in the wild.

How to use DevAgentRadar

• Read What the evidence shows as the only factual baseline—versions and behaviors are cited from the provided release signals, not rumor.
• Use For developers for personal toolchain moves this week; For teams for shared defaults, MCP allowlists, and cost/session policy.
• Skim What to ignore before planning upgrades; pin on GitHub-tagged stables when possible.
• Track Watch next and linked open proposals as the change backlog, not as shipped features.
• Re-run judgment when new versioned releases land—this brief is a snapshot of 12 signals / 22 tracked assistants in the current evidence window (≈55% GitHub/changelog-trusted mix).

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

• 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

• Cline desktop-v0.0.15: The app is now called Cline, renamed from Cline Code. · Your settings, sessions, and credentials carry over untouched — only the name and icon change · Refreshe — 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 v1.18.20: Surface failed subagent tool calls with a resumable task_id. · Retry provider responses that end with finish_reason: network_error. · Retry more network err — 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)