Skip to content
Guidance/Daily Briefing
digesteveryonedigestevidenceradarsecuritymodel11 min read · Oct 14, 2026

Radar brief — 2026-10-14

New-in-window GitHub/changelog signals first. mustKnow items are ranked first among today’s news; the ranking is not extra evidence. Several notes omit a large tail of changes (bodyOmitted) — those tails are not inferred below. Trusted mix is 54% GitHub/changelog vs HTML; Kiro a…

What the evidence shows

New-in-window GitHub/changelog signals first. mustKnow items are ranked first among today’s news; the ranking is not extra evidence. Several notes omit a large tail of changes (bodyOmitted) — those tails are not inferred below. Trusted mix is 54% GitHub/changelog vs HTML; Kiro and Factory Droid are HTML changelogs. One source reported errors (which source is not in this evidence). OpenAI Codex CLI rust-v0.163.0-alpha.11 has no changelog body.

Clinecli-v3.0.71
Key Signal & Impact
Must-know. Plugins load in-process in the hub; CLI no longer needs a JS runtime on PATH. Failed plugins warn by name, show as plugin_failure in cline config tools --json, and cline hub plugins lists status/last error (--reload re-imports one). CLINE_PLUGIN_MODE=sandbox keeps the old subprocess. --thinking now applies in --acp. Env API keys count as configured. Unchosen Cline provider model → first recommended (Claude Sonnet 5.5). Reasoning-off works on every provider; screenshots no longer inflate auto-compaction.
Clinev4.1.24
Key Signal & Impact
Must-know. Unpinned Cline and OpenRouter sessions use Claude Sonnet 5.5. Catalog refresh: Claude Haiku 5.5; AI/ML API, LowRouter, Prime Inference; Cline free list adds Step 5 Preview; v0 is only v0-1.5-md; OpenCode Go endpoint changes. Many providers change defaults (often Haiku 5.5 or Muse Wonder 1.0) if you do not pin a model. Reasoning-off 400s fixed; MCP/hook early-exit no longer crashes the extension host ~3s later; 404 on a retired model id is no longer shown as sign-in.
Clinesdk/sdk/v0.0.93
Key Signal & Impact
Same in-process plugin registry (timeouts, error capture, per-plugin state). session.create takes a plugins policy and returns pluginIssues. Each session gets its own plugin-module copy. Gemini/Vertex thinking-budget handling tightened.
Claude Codev2.1.299
Key Signal & Impact
@ file mentions can be path-only via CLAUDE_CODE_DISABLE_FILE_MENTIONS. Terminal idle compaction: 200K+ token chats idle ~54 minutes compact before the 1-hour prompt cache expires. Tune with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS or "idleCompaction": false (needs v2.1.290+). Managed-settings mods get $.prompt.media and $.session.identity(); agent view adds /effort and /fast. +77 changes omitted.
Google Antigravity1.3.5
Key Signal & Impact
Auto mode auto-approves file edits and safe commands, still prompts on irreversible work; commands that would leave the terminal sandbox are reviewed by a separate agent. --mode auto, /config → Agent Mode, or Shift+Tab. Linux/macOS only; Windows later. /compact on demand; /diff gets highlighting and folding.
Google Antigravity1.3.4
Key Signal & Impact
agy plugin list --json lists all installed plugins (not only imported). --model / /model can omit --effort and take the model default. Remote Control uses/saves real settings.json. Pay-as-you-go Business hides inaccessible third-party models. Failed-request retry cap 30s (was up to 4 min).
GitHub Copilot CLIv1.0.97
Key Signal & Impact
Agent-scoped declarative hooks for custom agents. Marketplace/plugin fetches retry once with the signed-in GitHub token. Autopilot denies a preToolUse "ask" instead of blocking on an unwatched dialog. Invalid config.json entry no longer wipes saved settings. Managed permission rules in /config are read-only/hidden unless the org set them. Custom-agent discovery beyond 100 / deep nests; /usage shows billed AI Credits.
GitHub Copilot CLIv1.0.96
Key Signal & Impact
Interactive sandbox settings suggest env secrets and masking hosts. /add-dir grants sandbox access for the session. Model IDs case-insensitive. Timeline shows who decided each permission (you, Assisted Permissions, policy, or unattended fallback). /allow-all stays available while enterprise policy is still resolving.
GitHub Copilot CLIv1.0.98-2
Key Signal & Impact
Prerelease. /ask expand/copy/reopen; command hooks complete when child-exit signals are lost under Node.
Kiro2.31.0 (CLI)
Key Signal & Impact
Local command sandbox + local autonomous mode for every V3 user. /sandbox contains agent shell (and optional local stdio MCP); agent, file tools, and remote MCP stay outside. /autonomous on switches to Default agent and turns sandbox on first. MCP tools load on demand; V3 config/state under KIRO_HOME. HTML changelog.
Kiro1.3.0 (IDE)
Key Signal & Impact
Same sandbox model; start saved Workflows from chat (/workflow-run); live session status on editor tabs. HTML changelog. Both Kiro notes warn the sandbox is not complete isolation.
Kilo Codev7.8.9
Key Signal & Impact
Pre-release. Drag-and-drop project order in Agent Manager (persists). Shift-click chat links to the system browser; split agent/model/reasoning control. +48 omitted.
Kilo Codejetbrains/v7.1.10-rc.3
Key Signal & Impact
Prerelease. Glob allowlist of ignored paths; Windows background process tracking via job object. Bundled CLI pin remains v7.8.9.
Zed AIv1.23.3
Key Signal & Impact
Stable: markdown-preview font list, edit-prediction worktree leak, remote-projects picker, Node v24 LTS, Threads Sidebar theme, opt-in checkpointing telemetry in the agent panel.
Zed AIv1.24.2-pre
Key Signal & Impact
Prerelease of the same class of fixes, plus a tab-indented code-block crash fix.
Factory Droidv0.240.0
Key Signal & Impact
Droid doctor can check model connectivity, MCP, and an opt-in full turn. Background commands stop when the starting Droid process exits; failed MCP servers can recover. App: Extra Usage one-click, remembered composer mode/autonomy over org session defaults. HTML changelog; +2 omitted.
Augment Codev1.2.0-prerelease.202610131516
Key Signal & Impact
Installer/SEA packaging notes (unsigned macOS -sea; Linux SHA256SUMS-sea). No product-behavior changelog in the evidence.
OpenAI Codex CLIrust-v0.163.0-alpha.11
Key Signal & Impact
Tag only: “Release 0.163.0-alpha.11”. No features or fixes to cite.

For developers

  • Cline (do this today): If you do not pin a model, expect Claude Sonnet 5.5 (Cline/OpenRouter) and different defaults on 302.AI, Cortecs, GitHub Copilot, Vertex, Novita, OpenCode Go, and others listed in v4.1.24. After cli-v3.0.71, run cline hub plugins (and cline config tools --json) instead of assuming a missing tool is “not installed.” Keep CLINE_PLUGIN_MODE=sandbox if you still want subprocess isolation. ACP users: --thinking finally applies.
  • Claude Code: Long idle terminal sessions (200K+ tokens, ~54 minutes) will compact before cache expiry — next message should not replay the whole transcript. Disable with "idleCompaction": false or raise CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS. Use CLAUDE_CODE_DISABLE_FILE_MENTIONS if @ should pass paths, not file bytes. Managed-settings mod authors: $.prompt.media and $.session.identity().
  • Antigravity: On Linux/macOS, try Auto mode (Shift+Tab / --mode auto) only if you accept auto-approved edits and “safe” commands. Use /compact before the window fills. --model name no longer requires --effort. If Remote Control in the browser used to be read-only, it now reads and writes settings.json.
  • Copilot CLI: Update through v1.0.97 for custom-agent hooks, config.json resilience, and unattended Autopilot that denies "ask" instead of hanging. Use /add-dir when the sandbox needs another tree; check the timeline for who approved each tool call.
  • Kiro: /sandbox then (only if you want unattended shells) /autonomous on. Review commands anyway — the notes say isolation is incomplete. Local stdio MCP can sit in the sandbox; remote MCP does not.
  • Zed: Take v1.23.3 for the crash/font/edit-prediction fixes. Checkpointing telemetry is opt-in and off when telemetry is disabled.
  • Kilo / Factory / Augment / Codex: Treat Kilo v7.8.9 and JetBrains rc.3, Augment prerelease, and Codex alpha.11 as non-stable. Factory v0.240.0 is useful if doctor/MCP recovery was failing for you.

For teams

  • Cline plugins are now in-process in the CLI/SDK. That removes a PATH/runtime dependency and a silent-failure mode, but it changes the isolation story. There is an open team proposal to keep CLINE_PLUGIN_MODE=sandbox. Decide the mode in managed environments before rolling cli-v3.0.71 / sdk v0.0.93.
  • Pin models in shared Cline/OpenRouter (and the long default-model list in v4.1.24). Unpinned seats will move to Sonnet 5.5 / Haiku 5.5 / other new defaults without a developer choosing it.
  • Copilot CLI org policy: managed permission rules are hidden/read-only in /config unless the org set them. Autopilot unattended runs will deny hook "ask" — hooks that expected a human prompt will not get one. Invalid config no longer resets trusted folders and sign-ins; still validate JSON in fleet images.
  • Antigravity Auto mode is a real permission change (auto file edits / safe commands; extra agent review only when leaving the terminal sandbox). Keep it off on Windows (not shipped) and in regulated repos until you review /config Agent Mode. Remote Control sessions now persist browser setting changes to disk.
  • Kiro sandbox does not sandbox the agent, built-in file tools, or remote MCP. Treat it as risk reduction plus /autonomous, not a compliance boundary. Confirm sandbox via the IDE status bar where that control exists.
  • Claude Code managed settings: $.session.identity() and $.prompt.media expand what org-installed mods can see (org/email when forceLoginOrgUUID matches; attached image/document bytes). Review mod install policy.
  • Zed: if you disabled telemetry, checkpointing telemetry does not apply; if you did not, it is now in the agent panel and still opt-in per the notes.

What to ignore

  • OpenAI Codex CLI rust-v0.163.0-alpha.11 — empty notes; do not infer behavior.
  • Augment Code v1.2.0-prerelease.* — installer/SEA trust metadata only; unsigned macOS -sea builds.
  • Prereleases: Copilot CLI v1.0.98-2, Kilo v7.8.9, Kilo JetBrains v7.1.10-rc.3, Zed v1.24.2-pre, Codex alpha.
  • Omitted tails (+77 Claude Code, +48 Kilo, +27 Cline SDK, etc.). Absence from this brief is not absence from the release.
  • Duplicate Zed rows — is the stable fix train; the -pre tag is not a second product event.
  • Kiro / Factory HTML changelogs — usable, but they are not GitHub release tags; Factory also omits 2 lines.
  • One error-reporting source in the mix — do not treat a single failing fetch as a product outage.

Watch next

  • Antigravity Windows Auto mode (explicitly “coming later”).
  • Whether Cline shops standardize on CLINE_PLUGIN_MODE=sandbox vs in-process (open team proposal, P5).
  • A real changelog for Codex 0.163.0 beyond the alpha.11 stub.
  • Copilot CLI v1.0.98 stable if the /ask + hook-signal fixes graduate from v1.0.98-2.
  • Open proposals also on the board (not evidence of shipping): re-pin Mistral Vibe thinking / Rust CLI; Devin Desktop ChatGPT sign-in + Fusion model pin; xAI Grok CLI harness/TUI; OpenCode PR #38252 verification videos.

How to use DevAgentRadar

DevAgentRadar is a free global radar for AI coding assistants. It ranks versioned evidence (GitHub releases and changelogs first) over marketing. isNew means the signal hit the radar in the last 36 hours — that is “today.” mustKnow is a judgment that daily users of that tool should see it first; always cite the release, not the badge. If bodyOmitted is set, stop at the published lines. Use For developers for personal pins, flags, and upgrade order; For teams for isolation, defaults, and org policy. Ignore empty tags, prereleases, and HTML-only notes until a versioned source repeats them. Track the open proposals if you want the next brief to measure those tools the same way.


Evidence appendix

  • New in the last 36 hours: 26 release(s) from 13 tool(s)
  • Evidence window: new releases first, then top curated signals (versioned releases preferred)
  • Assistants tracked: 29
  • Signals in this brief: 40
  • Trusted source mix (GitHub/changelog vs HTML): 54%
  • Open improvement proposals: 5
  • Sources reporting errors: 1

High-signal releases used

  • Claude Code v2.1.299: Added CLAUDE_CODE_DISABLE_FILE_MENTIONS environment variable: Claude gets the paths of @ file mentions to read with its file tools, not the files' contents — source

  • Cline sdk/sdk/v0.0.93: Plugins load in-process through one registry per process instead of a node/bun subprocess, so the compiled CLI and its hub no longer need a runtime on PATH, and — source

  • Google Antigravity 1.3.4: Added a --json flag to agy plugin list for machine-readable output, and fixed the command showing only imported plugins: it now lists every installed plugin — source

  • Kilo Code v7.8.9 — (pre-release): #14928 6522b4a - Re — source

  • GitHub Copilot CLI v1.0.97: 2026-10-13 · Add agent-scoped declarative hooks for custom agents · Remote plugin marketplace and plugin fetches retry once with your signed-in GitHub token aft — source

  • Cline cli-v3.0.71: Plugins now load inside the hub instead of a separate node or bun process, so the CLI no longer needs a JavaScript runtime on your PATH to run plugins. · A plug — source

  • Google Antigravity 1.3.5: Added Auto mode, which approves file edits and safe commands without asking and still prompts before irreversible actions; commands that would leave the termina — source

  • GitHub Copilot CLI v1.0.96: 2026-10-12 · Interactive sandbox settings suggest possible environment secrets and let you add masking hosts before saving. · Make model IDs in /model and /conf — source

  • Cline v4.1.24: With no model chosen, the Cline and OpenRouter providers now use Claude Sonnet 5.5. · For Cline, that's the first model in Cline's recommended list. · Selecting — source

  • Zed AI v1.24.2-pre: Fixed an issue where no fonts would show for the markdown preview configurations in the settings UI. (#65446 — source

  • Zed AI v1.23.3: Fixed an issue where no fonts would show for the markdown preview configurations in the settings UI. (#65447 — source

  • GitHub Copilot CLI v1.0.98-2: Improved · Expand /ask answers with Ctrl+E, copy the entire answer with Ctrl+Y, and reopen the last usable completed answer in the session with /ask or /btw — source

  • Kiro 2.31.0 — Local Command Sandbox and Local Autonomous Mode: October 13, 2026 Local Command Sandbox and Local Autonomous Mode · 2.31.0 · Use the local command sandbox and autonomous mode in V3, load MCP tools on demand, a — source

  • Kiro 1.3.0 — Local Command Sandbox, Start Saved Workflows from Chat, and Live Tab Status: October 13, 2026 Local Command Sandbox, Start Saved Workflows from Chat, and Live Tab Status · 1.3.0 · Limit what agent-run shell commands can reach, start save — source

Open proposals

Was this guidance useful?

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