Radar brief — 2026-09-02

What the evidence shows

GitHub-tagged releases dominate this window. Several HTML changelogs mix older patches into the same page; treat those as noisier. Large bodyOmitted tails mean the table only reflects listed notes, not the rest of the release.

For developers

• Cline first if tools vanished. Upgrade cli-v3.0.61 and SDK v0.0.82 before debugging prompts. Empty catalog capabilities were treated as an authoritative “no tools / no images” for Dify, SAP AI Core, opencode, and Codex CLI models.
• Lock a local run. Use Claude Code --restricted (or CLAUDE_CODE_RESTRICTED=1) when the job should stay inside the working tree with no shell, no WebFetch, and no bypassPermissions. Expect user/project/local settings files to be ignored.
• Operate Codex like a process manager. In rust-v0.149.0: codex agents, codex queue, /cd /pwd /cwd, and codex doctor. Confirm resumed/forked threads still have the permission profile you think they have. In rust-v0.152.0: package-style MCP names, per-tool output_token_limit, and rate-limit banners that deep-link to usage/credits/plans.
• Unstick OpenCode Claude 5 sessions. v1.18.26 keeps going after prompt or tool edits that used to fail on stale thinking blocks. Bedrock GPT-5.6 can now take none reasoning effort.
• Antigravity 1.1.23 is a reliability patch. Tab-complete /model; if subagents set enable_mcp_tools=true, this is the build that actually exposes the MCP dispatcher. Hook panics should no longer take down the CLI.
• Goose v1.48.0 is a provider mesh drop. Wire TrustedRouter / OpenCode Zen / OpenAI-compatible declarative providers only if you need those gateways. Budget against GLM-5.3 replacing GLM-5.2 and against custom cost fields actually driving tracking. Use on_failure on PreToolUse if you already run hooks.
• Zed v1.17.2: turn on low reasoning for DeepSeek V4 Flash/Pro when you want cheaper/faster passes; tabular previews are an IDE feature, not an agent control-plane change.
• Factory Droid v0.209.0: /archive current chats; new sessions with no model now land on GPT-5.6 Sol—set an explicit model if that is not what you want to pay for.

For teams

• Treat --restricted as a policy primitive, not a flag. Claude Code v2.1.248 can run without command/code tools, without WebFetch, without bypassPermissions, and without loading local settings. Use it for untrusted trees, CI, and contractor laptops. Pair with the new settings-load warnings in /doctor and /status (especially Bedrock/Vertex/custom ANTHROPIC_BASE_URL). Enterprise orgs billed via AWS Marketplace / self-serve / trials get /usage-credits.
• Do not roll Goose v1.48.0 from this brief alone. The listed provider/hook/cost notes are real; 98 further changes are omitted. Read the GitHub release before declaring TrustedRouter or OpenCode Zen a supported path. There is an open team proposal to register those providers.
• Cline Hub skew is now an ops event. cli-v3.0.61 prompts on an older running Hub, shows how many sessions a replace would interrupt, drains in-flight turns, and will not SIGKILL a wedged Hub. Windows binaries are Authenticode-signed; app-control blocks should print an actionable error. Fix remote MCP reachability (10s budget) so one dead SSE/HTTP server does not tear the session down.
• Codex permission and MCP defaults moved. v0.149.0 restores the saved permission profile on resume/fork instead of silently using current defaults. v0.152.0 keeps MCP tools across cache refreshes, allows package-style server names, and truncates tool output via output_token_limit—set that before a verbose server blows the context window. App-server thread/shellCommand timeouts can exceed one hour; that is a hang-risk as much as a feature.
• Default-model drift is a budget ticket. Factory Droid v0.209.0 defaults unset sessions to GPT-5.6 Sol. v0.208.1’s 60% GPT/Grok cut is explicitly limited-time. Windsurf v3.8.20 (HTML) adds an Enterprise “request more usage” link at ACU limits and warns on untrusted workspaces before local agents activate.
• Pin before you batch-upgrade. Open proposals already ask everyone to pin Cline models around cli-v3.0.61 and SDK v0.0.82. Those pins are not described in the release notes in this evidence window—verify on your catalog before a fleet bump.

What to ignore

• Kiro 2.20.2 — empty patch page; no actionable diff.
• Omitted tails — Goose +98, Codex 0.149 +151, Claude Code +43, Codex 0.152 +94, Zed +52. Do not infer extra MCP/security/breaking behavior from tags alone.
• HTML pages that concatenate history — Kiro 2.19.2 also reprints 2.19.1 and 1.0.337; Tabnine 6.5.4 reprints 6.5.3–6.5.1. Cite the version you actually installed.
• Factory’s 60% GPT/Grok discount — pricing promotion, not an agent capability.
• Proposal text that is not in the JSON notes — including any Cline “Claude Fable 5.1” default switch. Use the proposal as a ticket to verify, not as a release fact.
• One source in this window reported errors; if a changelog fetch looks thin (Kiro 2.20.2), assume the feed failed rather than that the product shipped nothing of consequence.

Watch next

• Goose remaining v1.48.0 notes (MCP, permissions, hooks, skills)—highest-score release with the largest omitted body.
• Whether Claude Code --restricted becomes the default for CI images, and how server-managed settings interact with ignored local files.
• Codex MCP output_token_limit plus package-style server names in real plugin graphs; 0.149.0 → 0.152.0 landed as two separate signals with no 0.150/0.151 in this brief.
• Cline open pins (cli-v3.0.61, SDK v0.0.82) after the capability-list fix.
• OpenCode v1.18.26 Claude 5 resume, Goose TrustedRouter / OpenCode Zen, Zed DeepSeek V4 low reasoning, Factory /archive + GPT-5.6 Sol default.
• Grok CLI proposals in the queue (prompt tag rename, legacy MCP / shadowed skills) even though Grok CLI itself has no release in this evidence set.

How to use DevAgentRadar

DevAgentRadar is a public radar, not a vendor blog. This brief is judgment over 15 curated signals across 23 tracked assistants in the current evidence window (about 49% GitHub/changelog vs HTML). Prefer the versioned GitHub rows above; open the source URL and read omitted bodies before you ship. Use For developers as personal upgrade actions and For teams as policy/default/spend tickets. The 8 open proposals are ready-made follow-ups—claim one rather than re-deriving the same changelog. If a page is empty, nested, or promotional, drop it in What to ignore and wait for the next versioned tag.

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

• Claude Code v2.1.248: Added --restricted (or CLAUDE_CODE_RESTRICTED=1): removes the built-in tools that run commands or code and WebFetch (unless named in --tools), keeps fil — 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

• Google Antigravity 1.1.23: Improved /model <name> autocompletion to accept the proposed model name ghost text with Tab. · Reduced subagent streaming overhead by sending subagent traje — source

• 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

• Zed AI v1.17.2: This week's release includes tabular data previews for CSV, TSV, PSV, and SSV files with sortable and resizable columns, value-based row filtering, and right-cl — 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 accept none reasoning effor — 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

• Windsurf v3.8.20: v3.8.20 August 21, 2026 Devin Desktop The Agent Command Center now follows whichever space is selected instead of being tied to one folder, converts an editor w — 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

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

Open proposals

• Upgrade OpenCode to v1.18.26 and resume Claude 5 sessions after prompt or tool edits (audience: developer, P5)
• Register TrustedRouter and OpenCode Zen as Goose declarative providers (audience: team, 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)
• Enable low reasoning effort for DeepSeek V4 Flash and V4 Pro (audience: developer, 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)
• Archive Factory Droid chats with /archive and default new sessions to GPT-5.6 Sol (audience: team, P3)