Skip to content
DevAgentRadarDevAgentRadar
Sign in
Guidance/Daily Briefing
digesteveryonedigestevidenceradarclimodel11 min read · Sep 11, 2026

Radar brief — 2026-09-11

GPT-6 Astra lands in Codex and Goose while CLI agents tighten worktrees, MCP, and permission sandboxes

What the evidence shows

Fifteen curated signals across 23 tracked assistants. Versioned GitHub releases carry the weight; HTML changelogs are noisier (trusted mix 49%). Themes cluster on cli, model, agent, mcp, and security, with four releases tagged breaking. Eight open improvement proposals are already filed from this window. One source in the mix reported errors.

GitHub / versioned releases (prefer these):

OpenAI Codex CLIrust-v0.154.0
Key Signal & Impact
GPT-6-Astra in the model picker and Amazon Bedrock catalogs. Experimental isolated checkouts via --worktree / /worktree. Inline answers while the agent keeps working. Windows can share a background Codex server. MCP OAuth refresh is coordinated (no auto-replay of rejected tools). Startup skips workspace-controlled helpers until trust; macOS sandbox blocks terminal input injection. Remote resume/fork keep saved permissions. (+147 changes omitted)
Goosev1.50.0
Key Signal & Impact
GPT-6 Astra support plus tool calling for goose-agent. HTTPS required for Snowflake. CLI session naming bound to a trusted path. Content-capture opt-out honored. Subagent platform guards, authoritative tool-approval details, permission revocations preserved across managers, latest MCP version preferred.
Claude Codev2.1.260
Key Signal & Impact
Fullscreen /diff panel for uncommitted edits. /cost and status line now explain likely prompt-cache misses. /reload-plugins and text /advisor work in headless/Desktop/SDK. Permission/sandbox fixes: path parentheses no longer drop Edit/Write/Read rules; a bad deny regex no longer breaks every edit; zsh REPORTTIME/REPORTMEMORY/DIRSTACKSIZE tricks no longer auto-approve. (+66 omitted)
Google Antigravity1.2.1
Key Signal & Impact
Custom agents can set excludeDefaultComponents: true and still keep post-invocation hooks. Transient 502/503/504/429 and mid-stream cuts retry in-process with backoff, keeping completed tool outputs. Open MCP/provider object schemas are no longer rejected. --continue no longer starts a new chat from a subdirectory/crash.
OpenAI Codex CLIrust-v0.153.0
Key Signal & Impact
Vim u / Ctrl+R undo-redo for full drafts. Plugin CLI can list/install/remove from remote marketplaces. tui.auto_recap = false keeps manual /recap. TUI reconnects after app-server drops, pausing uncertain submissions. Plus/Team get an earlier half-allowance warning. Full Access skips Guardian on confirmation-only actions. (+95 omitted)
Clinecli-v3.0.61
Key Signal & Impact
Stale Hub vs newer CLI now prompts (enter-to-replace / escape-to-keep) and drains in-flight turns. Windows binaries Authenticode-signed. Unreachable remote MCP gets a 10s budget instead of killing the session. Tool calling restored for Dify, SAP AI Core, opencode, and Codex CLI models whose empty capability lists had stripped every tool (and images). Langfuse tracing works in released builds.
Clinesdk/sdk/v0.0.82
Key Signal & Impact
Same capability-translator bug, fixed at the SDK: one shared translator so empty/["text"] catalogs no longer silently disable tools or vision. Langfuse detection is structural so minified builds actually trace.
OpenCodev1.18.26
Key Signal & Impact
Claude 5 sessions tolerate stale thinking blocks after prompt/tool changes. Bedrock GPT-5.6 accepts none reasoning effort; reasoning/replay handling tightened. apply_patch no longer emits an empty move path in permission metadata. Azure CLI sign-in asks for the resource name directly.
Zed AIv1.19.0-pre
Key Signal & Impact
Prerelease. Call hierarchy, Git Panel multi-select, untitled-buffer language detection. OpenRouter effort-based reasoning, better OpenAI-compatible streaming, lower memory on long agent sessions. (+50 omitted)

HTML / docs changelogs (lower confidence):

Windsurfv3.9.19
Key Signal & Impact
Devin Desktop Cascade is removed; Devin Local is the only Desktop agent. Migrate with Continue in Devin Local. Location picker is now Local / Worktree / Cloud. MCP registry fetches behind corporate proxies fixed. HTML changelog, not GitHub.
Factory Droidv0.209.0
Key Signal & Impact
/archive (Ctrl+X) from chat. Sessions with no model selected default to GPT-5.6 Sol. CLI starts even if org settings load slowly. Credit exhaustion suggests a live Droid Core model, not a retired one.
Factory Droidv0.208.1
Key Signal & Impact
Limited-time 60% off GPT and Grok, shown in the model selector. Pricing promo, not a capability change.
Tabninev6.5.4
Key Signal & Impact
Attribution feedback no longer dropped after VS Code diff / external-editor edits. Direct web fetch is default when tools.enableWebFetchTool is on (avoids LLM-mediated hallucination). Docs page, mixed with older 6.5.x notes.
Kirov2.20.2
Key Signal & Impact
Full-screen V3 /spec run with live progress and task-scope selection; Preserve scrollback toggle. Changelog also inlines unrelated 0.4.0 Crew notes (Windows/Linux installers, secrets vault). Duplicate thin patch signal exists for the same version. Treat as noisy.

For developers

  • Try GPT-6-Astra where it is actually wired: Codex CLI rust-v0.154.0 (picker + Bedrock) and Goose v1.50.0. Do not assume other assistants in this window shipped it.
  • Isolate agent checkouts. On Codex, --worktree / /worktree is experimental—use it for forks and parallel sessions, then browse/resume; do not treat it as a stable workflow yet. Windsurf’s picker now surfaces Local / Worktree / Cloud as first-class.
  • Upgrade Cline if tools or images vanished. cli-v3.0.61 and SDK v0.0.82 fix empty capability catalogs that silently stripped tools for Dify, SAP AI Core, opencode, and Codex CLI, and dropped images on file reads. Also upgrade if a dead SSE/HTTP MCP server was killing the TUI.
  • Use the new inspect surfaces. Claude Code v2.1.260: /diff beside the conversation; /cost + status prompt_cache for cache-miss causes; /reload-plugins and /advisor in headless. Codex: answer inline without losing the draft; /copy for status or session fields.
  • Custom agents: Antigravity 1.2.1 excludeDefaultComponents: true drops default prompt sections and built-in tools but keeps post-invocation hooks. OpenCode v1.18.26 is the one to grab if Claude 5 sessions were dying on stale thinking blocks, or if Bedrock GPT-5.6 needed none reasoning effort.
  • Do not ship Zed v1.19.0-pre as production. Memory and streaming improvements are real; it is still a prerelease.
  • Windsurf Desktop: Cascade is gone. Open existing threads with Continue in Devin Local or you are on a removed agent.
  • Factory: new sessions with no model selected now land on GPT-5.6 Sol; /archive is the session hygiene command. Confirm the model before a long run.
  • Tabnine: if tools.enableWebFetchTool is on, fetch is now direct—expect fewer hallucinated pages, and re-check attribution after diff acceptance (that path is what v6.5.4 fixed).

For teams

  • Treat permission and sandbox fixes as patch-now. Claude Code v2.1.260 closed three concrete bypass/breakage classes (parentheses in paths, uncompilable deny regexes taking down all edits, zsh assignment smuggling). Codex rust-v0.154.0 delays workspace-controlled helpers until trust and blocks macOS terminal input injection. Goose v1.50.0 requires HTTPS for Snowflake, binds session names to a trusted path, preserves permission revocations, and enforces subagent platform guards. Roll these before widening agent autonomy.
  • MCP is still the fragile edge. Codex coordinates OAuth refresh and surfaces login challenges instead of replaying rejected tools. Cline no longer dies on an unreachable remote MCP (10s budget). Antigravity now accepts open object schemas. Goose prefers the latest MCP version. Inventory remote MCP servers and OAuth apps after upgrade; a “quiet” catalog with empty capabilities is how Cline previously stripped tools.
  • Default-model and agent-product breaks need an explicit decision. Factory v0.209.0 defaults unset sessions to GPT-5.6 Sol. Windsurf v3.9.19 removes Cascade from Devin Desktop. Codex worktrees and Windows background daemons change how sessions are isolated and shared. Do not let these land as silent local upgrades.
  • Hub / CLI skew (Cline): a newer CLI against an old Hub now prompts with the number of sessions a replace would interrupt, drains in-flight turns, and will not kill a wedged Hub. Document the enter-to-replace vs escape-to-keep choice for anyone running a long-lived Hub.
  • Act on the open proposals from this window (do not invent extra scope):

What to ignore

  • Factory’s 60% GPT/Grok discount (v0.208.1): time-boxed billing, not a reason to change architecture.
  • Kiro’s changelog blob: v2.20.2 is mixed with 0.4.0 Crew installer/vault copy, plus a duplicate empty patch URL. Do not plan Windows/Linux “first-class” or secrets-vault work from that page without a cleaner source.
  • Zed v1.19.0-pre and omitted tails. Fifty-plus omitted Zed changes, 147 omitted Codex 0.154.0 changes, and similar tails elsewhere are not evidence. Do not infer features from bodyOmitted.
  • Tabnine plugin-version laundry lists and admin notes from v6.5.3/v6.5.2 that rode along on the v6.5.4 docs page.
  • Marketing language around “only agent” / “new default” until you confirm it on the versioned GitHub notes above. Windsurf’s Cascade removal is real; most other “defaults” need a pin, not a press mention.
  • The 1 error-reporting source is a quality flag on the mix, not a named incident in these 15 signals—do not invent an outage.

Watch next

  • Whether Codex worktrees leave experimental, and whether GPT-6-Astra quality holds in Bedrock vs the native picker (Codex rust-v0.154.0, Goose v1.50.0).
  • Goose-agent tool calling and subagent platform guards in real multi-agent runs.
  • Whether Cline’s capability-translator fix is sufficient, or whether the P4 pin-to-Fable proposals are still required after cli-v3.0.61 / SDK v0.0.82.
  • MCP OAuth refresh (Codex), open object schemas (Antigravity), and “latest MCP version” (Goose) colliding with corporate proxies (Windsurf’s HTML note).
  • Windsurf Desktop after Cascade removal: Devin Local migration completion, worktree/cloud picker adoption.
  • Factory defaulting unset sessions to GPT-5.6 Sol and the Droid Core suggestion when credits run out.
  • Zed v1.19.0 stable: keep the pre off the default IDE channel until a non-pre tag exists.
  • The two Grok CLI proposals (prompt tag rename; dropping legacy managed MCP / shadowed skills)—no matching Grok release in this brief.

How to use DevAgentRadar

DevAgentRadar is a free public radar for AI coding assistants. Judgment comes from versioned evidence, not vendor copy.

  1. Read What the evidence shows first. GitHub release tags beat HTML changelogs; omitted bodies are not secrets to fill in.
  2. Take For developers as local actions (upgrade, flag, pin, migrate your own sessions). Take For teams as policy: permissions, MCP, default models, Hub/CLI skew.
  3. Open the linked proposals when you want a tracked change with audience and priority—do not expand them past what the release notes actually say.
  4. Use What to ignore as a denylist for standup noise (promos, prereleases, mixed changelogs).
  5. Recheck Watch next on the following brief; 23 assistants move faster than any one vendor blog.

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

  • 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

  • Goose v1.50.0: Support GPT-6 Astra models #11869 · Tool calling for goose-agent [#11627](https://github.com/aaif-goose/goose/ — 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 1.2.1: Added support for excludeDefaultComponents: true in custom agent Markdown frontmatter, allowing custom agents to opt out of default prompt sections and built- — source

  • OpenAI Codex CLI rust-v0.153.0: Vim mode now supports undo with u and redo with Ctrl+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 — 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 accept none reasoning 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

Was this guidance useful?

Anonymous · one vote per visitor · re-click to clear. Helps improve digests—not a product ranking.