Radar brief — 2026-09-10
Codex CLI ships GPT-6-Astra plus experimental worktrees; Goose adds a declarative-provider wave; Claude Code and Cline patch permission and silent tool-calling bugs that change real agent behavior.
What the evidence shows
Fifteen curated signals across 23 tracked assistants. About half the mix is GitHub/changelog (49%); the rest is HTML/docs and is noisier. One source reported errors, so coverage is not complete. Several high-score GitHub releases omit long tails (bodyOmitted); only listed changes are treated as facts.
GitHub / versioned releases (preferred)
on_failure plus stable tool_call_id. Tagged breaking. 98 changes omitted.--worktree / /worktree isolated checkouts; inline answers while Codex keeps working; Windows shared background server; MCP OAuth refresh without auto-replaying rejected tools; trust-before-workspace-helpers and macOS sandbox blocking terminal input injection. Tagged breaking. 147 changes omitted./diff panel; /cost now explains likely prompt-cache misses; /reload-plugins and text /advisor in headless/Desktop/SDK; permission/sandbox fixes for paths with parentheses, uncompilable regex deny rules, and zsh assignments that hid command substitution. Tagged breaking. 66 changes omitted.503); faster sign-in from stored credentials; headless -p exits after the final answer, prints error: on stderr, and cuts idle latency. 11 changes omitted.tui.auto_recap = false; earlier Plus/Team usage warning; TUI reconnect after app-server drop; Full Access skips Guardian reviews for confirmation-only actions. 95 changes omitted.none reasoning effort; more reliable Bedrock reasoning/replay; apply_patch no longer emits an empty move path in permission metadata.HTML / docs changelogs (lower confidence)
/spec run) and a preserve-scrollback toggle. Changelog scrape also mixes older 0.4.0 “Crew” notes; treat as noisy. Duplicate near-empty signal ignored.tools.enableWebFetchTool is on (avoids LLM-mediated hallucination). Notes also bundle 6.5.3–6.5.1./archive (Ctrl+X) from chat; sessions with no model selected default to GPT-5.6 Sol; CLI starts even if org settings are slow; credit exhaustion suggests a live Droid Core model.For developers
- Upgrade Codex CLI to rust-v0.154.0 if you want GPT-6-Astra. Try
--worktree//worktreefor isolated new or forked sessions, then browse/resume those checkouts. Use inline answers so a question does not wipe the main draft. On Windows, expect a shared background server plus daemon lifecycle commands. - If you stayed on rust-v0.153.0, set
tui.auto_recap = falsewhen auto recaps get in the way; use the plugin CLI to list/install/remove marketplace plugins; confirm Vimu/Ctrl+Rkeeps attachments. - Goose v1.48.0 is a provider and hook release. Wire new declarative providers only if you need them; switch Z.ai callers from GLM-5.2 to GLM-5.3; use custom provider cost fields if you track spend; add
on_failureon PreToolUse hooks and rely on stabletool_call_idacross the tool lifecycle. - Claude Code v2.1.260: turn on
/diffin fullscreen while it edits; readprompt_cache//costwhen a cache miss looks expensive; use/reload-pluginsand/advisorin headless/SDK. Re-test deny rules whose paths contain()or broken character classes — one bad pattern used to fail every edit; parenthesis paths were dropped as invalid and left “read-only” folders writable. zshREPORTTIME/REPORTMEMORY/DIRSTACKSIZEtricks now prompt instead of auto-approving. - Cline users on Dify, SAP AI Core, opencode, or Codex CLI models should update CLI v3.0.61 and SDK v0.0.82. Empty capability lists had silently stripped tools and images. Also update if remote SSE/HTTP MCP servers are sometimes down (10s connect budget) or if Langfuse never showed data in production builds.
- OpenCode v1.18.26: resume Claude 5 threads that used to die on stale thinking blocks; send Bedrock GPT-5.6 with
nonereasoning effort if that is what you mean; re-checkapply_patchpermission metadata after the empty move-path fix. - Antigravity 1.1.28: keep headless
-pruns; fatal errors now hit stderr with a stableerror:marker, and brief503s should retry instead of killing the session. Bound wait time with--print-timeout; daemons such as dev servers are left running. - Zed v1.19.0-pre is a prerelease. Useful if you need OpenRouter effort-based reasoning or longer agent sessions, not a production pin.
- Factory Droid v0.209.0: new sessions with no model pin now land on GPT-5.6 Sol — set the model explicitly if that is not what you want. Use
/archiveto clear the chat list. - Tabnine v6.5.4: if you enabled
tools.enableWebFetchTool, fetches are now direct; verify attribution still records after VS Code diff acceptance.
For teams
- Treat Claude Code v2.1.260 as a security patch, not a UI release. Path-permission bugs could leave folders writable and auto-approve disguised zsh. Re-audit sandbox/deny rules, then roll the version org-wide.
- Codex rust-v0.154.0 hardens startup and MCP. Workspace-controlled helpers should not run before trust; macOS sandbox blocks terminal input injection; remote resume/fork keeps saved permissions; fresh sessions follow server model defaults unless overridden. rust-v0.153.0 “Full Access skips Guardian for confirmation-only actions” is a policy change — decide whether that matches your review bar.
- Cline Hub upgrades are now interactive. A newer CLI will tell you how many sessions a Hub replace would interrupt, drain in-flight turns, and refuse to kill a wedged Hub. Plan Windows rollout around Authenticode-signed binaries and application-control error text.
- Windsurf v3.9.19 removes Cascade from Devin Desktop. Devin Local is the only Desktop agent. Tell people to use Continue in Devin Local before old threads vanish from muscle memory. Location model is Local / Worktree / Cloud; MCP registry fetches behind corporate proxies are called out as fixed.
- Default-model drift: Factory Droid v0.209.0 defaults unset sessions to GPT-5.6 Sol; Goose routes GPT-5.6 (sol/terra/luna) on Bedrock; Codex 0.154.0 adds GPT-6-Astra. Pin models in team config if you need comparable evals or cost.
- Cost and promo noise: Goose can track cost from custom provider fields. Factory v0.208.1 advertises 60% off GPT and Grok for a limited time — do not bake that rate into budgets. Codex 0.153.0 warns Plus/Team users earlier in a ~five-hour usage window.
- Goose, Codex 0.154.0, Claude Code, and Zed pre are tagged breaking. Read the listed items before a fleet bump; do not assume the omitted tails are safe.
What to ignore
-
The second Kiro
2.20.2signal (title/date only) and the 0.4.0 “Crew” block mixed into the v2.20.2 HTML scrape. -
Tabnine notes for 6.5.3–6.5.1 bundled under the 6.5.4 heading, except where you still run those older plugin builds.
-
Long
+N more changestails on Goose, Codex, Claude Code, Zed, Cline, and Antigravity — they are omitted, not summarized here. -
Zed Guild marketing links and ChatGPT-subscription cancel-button noise in the Zed prerelease.
-
Factory’s 60% GPT/Grok discount as a durable pricing fact.
-
The OpenCode “PR #38252 verification videos” proposal (P2) unless you are actually reviewing that PR.
-
Treating Windsurf / Kiro / Tabnine / Factory HTML pages as equal to GitHub release tags. This window’s trusted mix is only 49%, and one source reported errors.
Watch next
- Open proposals (8): switch Codex CLI to GPT-6-Astra and isolated
--worktreecheckouts (everyone, P5); pin Cline models around CLI v3.0.61 and SDK v0.0.82 (everyone, P4); Grok CLI prompt/MCP cleanup (developer, P4); continue Cascade threads in Devin Local on Windsurf 3.9.19 (everyone, P3); archive Factory Droid chats and default new sessions to GPT-5.6 Sol (team, P3). - Whether GPT-6-Astra (Codex) or GPT-5.6 Sol (Factory/Goose/OpenCode Bedrock) becomes the practical default in CLI agents.
- Worktrees as a cross-tool pattern (Codex experimental worktrees; Windsurf Local / Worktree / Cloud).
- MCP auth and fleet behavior: Codex OAuth refresh, Cline unreachable-server timeout, Windsurf proxy registry fetches, Goose provider declarations.
- Zed v1.19.0 stable, if/when it graduates from
-pre. - Claude Code permission/sandbox follow-ups after the parentheses, regex, and zsh-substitution fixes.
How to use DevAgentRadar
DevAgentRadar is a free public radar for AI coding assistants: judgment from versioned evidence, not vendor copy. This brief is one evidence window (15 signals, 23 assistants). Prefer GitHub release tags over HTML changelogs; ignore omitted bodies; treat prereleases and mixed scrapes as low confidence. Developers should pin models, re-test permission/MCP paths, and try isolation flags on their own machine. Teams should map breaking/security tags to rollout policy and use the open proposals as concrete change tickets, not as extra facts beyond the releases above.
Evidence appendix
- Evidence window: top curated signals (versioned releases preferred)
- Assistants tracked: 23
- Signals in this brief: 15
- Trusted source mix (GitHub/changelog vs HTML): 49%
- Open improvement proposals: 8
- Sources reporting errors: 1
High-signal releases used
-
Goose
v1.48.0: TrustedRouter declarative provider #11422 · OpenCode Zen gateway as a declarative provider [#10939](https://gi — source -
OpenAI Codex CLI
rust-v0.154.0: GPT-6-Astra is now available in the model picker and Amazon Bedrock catalogs. (#42879, #42619) · Experimental worktree support lets you create isolated checkout — source -
Claude Code
v2.1.260: Added a diff panel that opens beside the conversation in fullscreen mode and shows your uncommitted changes as Claude edits; toggle it with/diff· Added a li — source -
Zed AI
v1.19.0-pre: This week's release includes call hierarchy support, multi-select in the Git Panel, automatic language detection for untitled buffers, and project search on typ — source -
Google Antigravity
1.1.28: Improved resilience to transient model API errors: the agent now retries errors such as503 Unavailablefor much longer with exponential backoff, so a brief s — source -
OpenAI Codex CLI
rust-v0.153.0: Vim mode now supports undo withuand redo withCtrl+R, preserving complete drafts including pasted content and attachments. ([#41941](https://github.com/op — source -
Cline
cli-v3.0.61: Cline now handles a running Hub that is older than your CLI. · Instead of quietly talking to a hub executing stale code, you get a prompt showing how many activ — source -
Cline
sdk/sdk/v0.0.82— SDK v0.0.82: Fixed tool calling being silently disabled for gateway models whose catalog entry declares no capabilities. · Three separate producers built gateway model defin — source -
OpenCode
v1.18.26: Claude 5 sessions now tolerate stale thinking blocks instead of failing after prompt or tool changes. · Bedrock GPT-5.6 models now acceptnonereasoning effor — source -
Windsurf
v3.9.19: v3.9.19 September 8, 2026 Devin Desktop Cascade has been removed. · Devin Local is now the only agent available in Devin Desktop. · Use the Continue in Devin Lo — source -
Kiro
v2.20.2: 2.20.2 Aug 31, 2026 · 2.20.1 Aug 27, 2026 Full-Screen Spec Task Execution and Preserve Scrollback This release gives V3 spec runs a full-screen task execution v — source -
Tabnine
v6.5.4: v6.5.4 August 30, 2026 Bug Fixes Tabnine CLI: Provenance & Attribution : Fixed attribution feedback being dropped after tool inputs were edited via VS Code diff — source -
Factory Droid
v0.209.0— · Desktop v0.166.0 (September 1, 2026): Session archiving from chat, a new default model, faster startup, and steadier app views · Archive sessions from chat - Archive the current session with /ar — source -
Factory Droid
v0.208.1— · Desktop v0.165.1 (August 29, 2026): 60% off GPT and Grok models · GPT and Grok discount - GPT and Grok models now bill at 60% off for a limited time, with the discount shown in the model selec — source
Open proposals
- Switch Codex CLI to GPT-6-Astra and isolated
--worktreecheckouts (audience: everyone, P5) - Pin Cline models after the v3.0.61 switch to Claude Fable 5.1 (audience: everyone, P4)
- Pin Cline provider models before SDK v0.0.82 switches 57 defaults to Claude Fable 5.1 (audience: everyone, P4)
- Replace <output_efficiency> with <response_guidelines> in Grok CLI prompts (audience: developer, P4)
- Drop legacy managed MCP configs and shadowed platform skills in Grok CLI (audience: developer, P4)
- Continue Cascade conversations in Devin Local on Desktop 3.9.19 (audience: everyone, P3)
- Archive Factory Droid chats with /archive and default new sessions to GPT-5.6 Sol (audience: team, P3)
- Improve with OpenCode pr-38252-videos: pr-38252-videos PR #38252 verification videos (audience: everyone, P2)
Anonymous · one vote per visitor · re-click to clear. Helps improve digests—not a product ranking.