Skip to content
Guidance/Daily Briefing
digesteveryonedigestevidenceradarmodelcli11 min read · Sep 7, 2026

Radar brief — 2026-09-07

Permission and sandbox holes close in Claude Code while Goose floods the provider catalog, Codex ships a plugin CLI, and several agents quietly restore tools, MCP auth, and stream recovery.

What the evidence shows

Window: 14 signals across 23 tracked assistants. Versioned GitHub releases dominate the useful rows; the mix is only 49% GitHub/changelog vs HTML, with 1 source reporting errors and 8 open improvement proposals. Several high-score notes omit dozens of extra commits (bodyOmitted) — those are not inferred below. Zed v1.19.0-pre is a prerelease. Kiro 2.20.2 published a date stamp and almost no change text.

Goosev1.48.0
Key Signal & Impact
Declarative providers: TrustedRouter, OpenCode Zen gateway, Gondola, SayGM, Lynkr, PleumRouter. Z.ai GLM-5.2 replaced by GLM-5.3. Custom provider cost fields feed cost tracking. Model-native audio transcription. GPT-5.6 (sol/terra/luna) via AWS Bedrock; GPT-5.6 follow-up for Codex and Responses API. PreToolUse on_failure, PreToolUseResult, stable tool_call_id. Saved recipes selectable when creating a schedule. 98 further changes omitted. Tagged breaking (hooks/MCP/permissions).
Claude Codev2.1.260
Key Signal & Impact
Fullscreen /diff panel of uncommitted edits. /cost and status-line prompt_cache now show a likely cache-miss cause. /reload-plugins and text /advisor in headless/Desktop/SDK. Gateway: oidc.scope_on_refresh; desktop policy keys userPluginMarketplacesEnabled / userPluginUploadsEnabled. Security: path parentheses no longer drop Edit/Write/Read rules (read-only folders were left writable); a single uncompilable deny regex no longer breaks every edit; zsh REPORTTIME/REPORTMEMORY/DIRSTACKSIZE assignment smuggling no longer auto-approves. 66 further changes omitted. Tagged breaking (permissions/sandbox/hooks/compaction).
Zed AIv1.19.0-pre
Key Signal & Impact
Editor: call hierarchy, Git Panel multi-select, untitled-buffer language detection, project search-on-type, LSP 3.18 Markdown diagnostics. Agent: OpenRouter effort-based reasoning, OpenAI-compatible streaming, lower memory on long Agent sessions, rename Terminal Threads, cancel ChatGPT subscription setup. Prerelease. 50 further changes omitted.
Google Antigravity1.1.25
Key Signal & Impact
Opt-in workspace-grouped /resume (Ctrl+F). Gemini 3.8 Flash in catalog with GEMINI_API_KEY. Markdown custom agents inherit ambient skills, rules, and subagents by default. MCP OAuth fixed for authorization codes >1024 chars. Windows /skills path grouping; relative subagent/plugin paths resolve from the agent file. Duplicate permission grants no longer accumulate. /boost worker-tool fix; nil-pointer crash after trajectory close; Remote Control reverse-tunnel hardened.
OpenAI Codex CLIrust-v0.153.0
Key Signal & Impact
Vim undo u / redo Ctrl+R (drafts, pastes, attachments). Plugin CLI: list/install/remove from remote marketplaces. tui.auto_recap = false keeps manual /recap. TUI history shows full patches, background-terminal input, completed commands. Plus/Team: earlier warning when < half of a ~5-hour allowance remains. TUI reconnects after app-server drop (drafts/transcripts kept; queued submits paused). Full Access skips Guardian reviews for confirmation-only actions. 95 further changes omitted.
OpenAI Codex CLIrust-v0.152.0
Key Signal & Impact
Vim / ? draft search (n/N). Rate-limit banners (usage, credits, limits, plans). Credential-refresh progress including Bedrock. MCP server names may include : @ / .. Per-tool output_token_limit with consistent resume truncation. App-server thread/shellCommand timeouts (including >1 hour). Auto-approval reviews keep longer transcripts and user instructions across compaction. MCP tools survive cache refresh; auth retries use refreshed helper headers. 94 further changes omitted.
Clinecli-v3.0.61
Key Signal & Impact
Newer CLI vs older Hub: prompt with interrupted-session count; drain in-flight turns; do not kill a wedged Hub. Windows binaries Authenticode-signed (Azure Trusted Signing) with actionable WDAC errors. Unreachable remote MCP no longer kills the CLI (10s connect budget). Tool calling was silently disabled for Dify, SAP AI Core, opencode, and Codex CLI catalog entries with empty capabilities; images were dropped the same way. Langfuse tracing works in released/minified builds. 18 further changes omitted.
Clinesdk/sdk/v0.0.82
Key Signal & Impact
Same capability bug, root cause: three drifted translators; builtin path emitted ["text"] vs undefined, read as an authoritative denial. One shared translator. Empty capability lists no longer strip vision/attachments or lie in model pickers. Langfuse detection is structural (minification-safe) and moved to AI SDK 7 telemetry. 34 further changes omitted.
OpenCodev1.18.29
Key Signal & Impact
Codex OAuth model filter accepts integer GPT versions such as gpt-6 (gpt-6-astra now appears for OpenAI subscription users). Console quota-reset action. Small, targeted patch.
Kirov2.19.2
Key Signal & Impact
HTML changelog. Spec-review mouse support (m to toggle). Stream idle watchdog (warn 60s / cancel 300s), retries with backoff, default 60-minute stream timeout; tunables api.streamIdleSoftTimeout, api.streamIdleHardTimeout, api.timeout. Body also mixes older IDE notes (e.g. 1.0.337). 8 further changes omitted.
Tabninev6.5.4
Key Signal & Impact
Docs changelog (also rolls 6.5.3–6.5.1). CLI provenance/attribution no longer dropped after VS Code diff or external-editor edits. Direct web fetch is default when tools.enableWebFetchTool is on (avoids LLM-mediated hallucinated fetch). CLI 0.35.0. 6.5.3: org-level repos; per-team model assignment; Plan Mode save fix. 6.5.2: WebStorm 2026.2 startup crash.
Factory Droidv0.209.0
Key Signal & Impact
Docs changelog (+ Desktop v0.166.0). /archive or Ctrl+X; restore from Archived. Unset sessions default to GPT-5.6 Sol. CLI starts even if org settings are slow. App: diff selection, environment panel, file-link line targets. Out-of-credits suggests a living Droid Core model.
Factory Droidv0.208.1
Key Signal & Impact
Limited-time 60% off GPT and Grok billing, shown in the model selector. Promo, not a runtime change.
Kiro2.20.2
Key Signal & Impact
HTML patch page dated Aug 31, 2026. No change body in this signal — do not treat as a feature release.

For developers

  • Patch Claude Code to v2.1.260 if deny/allow paths contain () or brittle globs, or if you run zsh. Those were permission/sandbox bugs, not cosmetics. Use /diff in fullscreen to watch uncommitted edits; use /cost when prompt_cache looks empty (tool defs, system prompt, or TTL idle).
  • Patch Cline CLI v3.0.61 and SDK v0.0.82 if you route through Dify, SAP AI Core, opencode, or Codex CLI. Empty catalog capabilities were stripping all tools and images. Unreachable SSE/HTTP MCP should no longer tear down the TUI; expect a 10s connect budget instead.
  • Goose v1.48.0: if you live on Z.ai, confirm GLM-5.3; if you bill custom OpenAI-compatible endpoints, fill the new cost fields. PreToolUse hooks can use on_failure; tool_call_id is now stable across the tool lifecycle. Treat the release as breaking until you read the omitted 98 items yourself.
  • Codex CLI rust-v0.152.0 then rust-v0.153.0: package-style MCP names (: @ / .), per-tool output_token_limit, long thread/shellCommand deadlines, then plugin marketplace CLI + vim undo/search + tui.auto_recap = false. After an app-server drop, review paused queued submits before they resume.
  • Antigravity 1.1.25: Markdown custom agents now inherit ambient skills/rules/subagents — pin or isolate agent files if you wanted a blank agent. MCP OAuth with long codes should work; re-check Windows /skills grouping and relative plugin paths.
  • OpenCode v1.18.29: subscription users who expected gpt-6-astra should refresh the OAuth model list.
  • Tabnine: upgrade CLI to 0.35.0; turn on tools.enableWebFetchTool if you want direct fetch instead of an LLM-mediated page summary. Related proposal: Enable tools.enableWebFetchTool and upgrade Tabnine CLI to 0.35.0.
  • Kiro v2.19.2: tune stream idle/hard timeouts if long generations were dying; do not wait on 2.20.2 for a described fix.
  • Do not ship Zed v1.19.0-pre as production IDE policy; memory/streaming gains are real but the tag is prerelease.
  • Open Grok CLI proposals (not in this release table): replace <output_efficiency> with <response_guidelines> and drop legacy managed MCP configs / shadowed platform skills.

For teams

  • Treat Claude Code v2.1.260 as a security patch. Parentheses in path rules could leave “read-only” trees writable; zsh assignment smuggling could auto-approve. Roll it before widening sandbox trust. If your IdP only returns id_token on refresh when openid is re-requested, set oidc.scope_on_refresh. Desktop policy now has explicit plugin marketplace/upload keys — decide those org-wide, then use /reload-plugins in headless/SDK.
  • Cline Hub upgrades are no longer silent. Operators get a session-count prompt, drain in-flight turns, and will not SIGKILL a wedged old Hub. Windows: Authenticode via Azure Trusted Signing; expect clearer application-control failures. Review Pin Cline models after the v3.0.61 switch to Claude Fable 5.1 before a fleet default drifts. Plugin-layout proposal (team, P5): move Agent Plugins into ~/.agents/plugins and Hub-enable them.
  • Codex Plus/Team: warn users about the half-allowance banner in the ~5-hour window. Full Access skipping Guardian on confirmation-only actions is a control-plane change — confirm that matches your review policy. Remote plugin install/remove is now a first-class CLI.
  • Goose: provider catalog and Bedrock GPT-5.6 routing expand shadow-IT surface; use the new cost fields so custom providers are not free in dashboards. Hooks/on_failure belong in the same change-control as MCP permissions.
  • Antigravity: duplicate permission grants were accumulating across reloads and subagents — audit configs after upgrade. Remote Control reverse-tunnel routing was hardened; re-verify OAuth IdPs that emit long authorization codes.
  • Factory Droid: new sessions with no model now default to GPT-5.6 Sol; out-of-credits messaging no longer points at a retired Core model. /archive is the team hygiene action — see Archive Factory Droid chats with /archive and default new sessions to GPT-5.6 Sol. Treat v0.208.1 60% off GPT/Grok as a billing promo with an expiry, not a capability upgrade. CLI no longer blocks startup on slow org settings — settings apply late; document that race for policy-sensitive flags.
  • Tabnine 6.5.3: org-level repository add no longer depends on which team is selected; verify per-team model assignment actually applies.

What to ignore

  • Kiro 2.20.2 — date-only HTML patch, no evidenced changes.
  • Kiro v2.19.2 body mixing 1.0.337 IDE notes with 2.19.x stream recovery. Use the named 2.19.1/2.19.2 items only.
  • Factory Droid v0.208.1 discount copy and any “+N more changes” you cannot see (bodyOmitted on Goose 98, Claude Code 66, Zed 50, Codex 95+94, Cline 18+34, Tabnine 7, Kiro 8).
  • Zed v1.19.0-pre as a stable team standard.
  • Conflating OpenCode v1.18.29 with Goose’s “OpenCode Zen gateway” provider — different repos, different changes.
  • Windsurf agent-window and OpenCode PR-video proposals in this brief’s release table; they are open proposals, not shipped evidence here.
  • Marketing language, plugin-version laundry lists, and internal scoring. HTML/docs rows (Kiro, Tabnine, Factory) are noisier than GitHub tags.

Watch next

  • The omitted tails on Goose v1.48.0 and Claude Code v2.1.260 (both tagged breaking; hooks, compaction, permissions, MCP).
  • Whether GPT-5.6 routing stays consistent: Goose Bedrock sol/terra/luna, Factory default Sol, OpenCode gpt-6* OAuth filter — three different surfaces, not one guarantee.
  • Plugin marketplaces becoming a real control plane: Codex remote install/remove, Claude headless /reload-plugins + desktop marketplace keys, Cline Hub plugin-path proposal.
  • MCP auth/names: Antigravity long OAuth codes, Codex package-style server names and output_token_limit, Cline 10s remote-MCP budget.
  • Cline Fable 5.1 pin (P4, everyone) and whether empty-capability catalogs regress again on other gateways.
  • Zed 1.19.0 stable follow-through on Agent memory and OpenRouter effort.
  • Remaining open proposals (8), including Grok CLI MCP/prompt cleanup and the low-priority OpenCode verification-videos item.

How to use DevAgentRadar

DevAgentRadar is a free global radar for AI coding assistants: we score versioned evidence (prefer GitHub releases/changelogs) and refuse to invent omitted commits or unreleased products.

  1. Read What the evidence shows as the source of truth — table rows only, no vendor blogs required.
  2. Take For developers as personal upgrade/config actions; take For teams as policy, permissions, billing, and fleet defaults.
  3. Skip What to ignore so HTML stubs, prereleases, and truncated +N more lines do not drive purchases.
  4. Track Watch next and the open proposals (8 in this window) as the actionable backlog — each proposal is already audience-tagged (developer / team / everyone).
  5. When GitHub share is ~50% as it is today, discount HTML docs (Kiro, Tabnine, Factory) until a tag confirms the same fact.

Re-run against the live feed before you pin defaults: this brief is a snapshot of 14 curated signals, not a complete catalog of all 23 tracked assistants.


Evidence appendix

  • Evidence window: top curated signals (versioned releases preferred)
  • Assistants tracked: 23
  • Signals in this brief: 14
  • 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://gisource

  • 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 1.1.25: Added an opt-in workspace-grouped view to the /resume conversation picker, allowing users to toggle between a flat list and conversations grouped by directory — 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/opsource

  • OpenAI Codex CLI rust-v0.152.0: Vim mode supports / and ? searches within drafts, highlighted matches, and repeat navigation with n and N. (#41586) · Rate-limit banners offer actions f — 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 v1.18.29: Allow Codex OAuth model filtering to recognize integer GPT versions like gpt-6 · Fixes issue of gpt-6-astra not showing up for openai subscription users · **T — source

  • Kiro v2.19.2: 2.19.2 Aug 25, 2026 · 2.19.1 Aug 21, 2026 Spec Review Mouse Support and Automatic Stream Recovery This release brings mouse support to the spec review screen. · — 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

  • Kiro 2.20.2 — Aug 31, 2026: 2.20.2 Aug 31, 2026 — source

Open proposals

Was this guidance useful?

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