Skip to content
Guidance/Daily Briefing
digesteveryonedigestevidenceradarmodelagent8 min read · Aug 15, 2026

Radar brief — 2026-08-15

Agent CLIs tighten plan-mode guardrails, plugin catalogs, and spend attribution—while enterprise sign-in and safer auto-approvals move from nice-to-have to release-note defaults.

What the evidence shows

Versioned GitHub-style releases dominate this window (16 signals across tracked assistants). Strongest clusters: agent loop / session control, MCP & plugins, CLI approvals & sandbox posture, and security defaults.

Claude Codev2.1.233
Key Signal & Impact
GitLab MR URLs in --worktree / agents view (!N); opt-in forward_user_identity on Anthropic upstreams for per-user spend attribution; opt-in Linux Bash memory cgroup (CLAUDE_CODE_TOOL_MEMORY_LIMIT) to contain runaway builds
OpenAI Codex CLIrust-v0.147.0
Key Signal & Impact
Portable Agent Plugins + multi-catalog search; persistent ordered conversation sections; --approve-for-me auto-reviewed approvals; import/sync Cursor & Claude skills/conversations without dupes
OpenAI Codex CLIrust-v0.146.1
Key Signal & Impact
Safer automatic-review defaults for cyber-capable models; permission changes explained in the terminal UI
Zed AIv1.15.0
Key Signal & Impact
git.diff_base chooses gutter/status/git: diff against HEAD vs default-branch merge base; Project Panel drag-out on macOS/Linux Wayland; JSX/TSX linked editing + Emmet in return/arrow bodies
Google Antigravity1.1.10
Key Signal & Impact
Business sign-in for Gemini Enterprise (Cloud project auth, license seats, regional inference, org admin controls)
Goosev1.46.0
Key Signal & Impact
Unrolled agent loop; cache-safe append-only turn context; streaming shell output; per-message usage/cost/TTFT/tok-s UI and session totals
Clinedesktop-v0.0.8
Key Signal & Impact
Edit any earlier message → fork + checkpoint rewind + re-run; transactional workspace-atomic restores; more reliable checkpoints (restart/compaction/resume)
Clinev4.1.4
Key Signal & Impact
Skills beside workflows in slash menu (no silent shadowing); remove model-initiated plan→act; hard-block file-editing shell in plan mode (read-only investigate still allowed)
Clinesdk/sdk/v0.0.72
Key Signal & Impact
Queued prompts survive aborts; consistent full-stop; durable session context across aborts/hub restarts; failed queued turns → run.failed
Gemini CLIv0.55.1
Key Signal & Impact
Security: case-insensitive sensitive path blocklist + VS Code HITL; tool registry discovery; defensive at-reference path resolution; release/CI hardening
OpenCodev1.18.17
Key Signal & Impact
Compaction keeps complete recent turns + clearer small-model summaries; retry cap + jitter; PDF attachments where Copilot advertises vision; model routing/sampling fixes (DeepSeek V4 Flash, Muse, MERGE Gateway reasoning)
Windsurfv3.7.25
Key Signal & Impact
Devin Desktop sidebar perf for huge session caches; MCP auth fixes (self-hosted GitLab, older Atlassian MCP)—HTML changelog, not GitHub tag
xAI Grok CLI (Grok Build)be71313
Key Signal & Impact
Monorepo sync: presence protocol E2E; worktree fetch tightening; pager/tool-card UX; CLI update telemetry—commit signal, not a semver release

Lower-confidence / noisy in this set: Gemini v0.56.0-preview.1 and nightly v0.56.0-nightly.20260815.* (prerelease/caretaker-heavy); Codex rust-v0.148.0-alpha.19 (tag-only alpha). Treat as watchlist, not adopt signals.

For developers

  • Plan mode is becoming enforcement, not vibes. On Cline v4.1.4, expect hard blocks on mutating shell (editors, redirects, git writers, installs) while investigating stays read-only—and you leave plan mode, not the model. Rehearse workflows that assumed model-driven plan→act.
  • Try plugin/skill portability before re-prompting the same setup. Codex rust-v0.147.0 installs portable Agent Plugins and can import Cursor-managed skills; Cline surfaces skills next to workflows with name disambiguation. Inventory local skills once, then sync deliberately.
  • Use session fork/rewind when experimenting. Cline Desktop v0.0.8 lets you edit an earlier turn, restore the workspace checkpoint atomically, and re-run—prefer that over manual git stash archaeology mid-agent.
  • Contain runaway local tools. Claude Code v2.1.233: consider opt-in CLAUDE_CODE_TOOL_MEMORY_LIMIT on Linux if Bash builds have wedged sessions.
  • Diff gutters with intent. Zed v1.15.0 git.diff_base—set explicitly if you care about “uncommitted vs HEAD” vs “branch vs default merge base.”
  • Read security patches on CLI upgrades. Gemini CLI v0.55.1 tightens sensitive-path blocklisting (case-insensitive) and path resolution; Codex rust-v0.146.1 tightens auto-review defaults for cyber-capable models—don’t casually loosen approvals to “match old behavior.”
  • Prefer stable tags over nightlies/alphas for daily drivers (skip Gemini 0.56 previews/nightlies and Codex 0.148 alpha unless you are validating a fix).

For teams

  • Spend and identity attribution: Claude Code v2.1.233 opt-in forward_user_identity on Anthropic upstreams—pilot behind your apps gateway if you need per-user cost allocation through a proxy.
  • Enterprise auth path: Google Antigravity 1.1.10 Business sign-in (Gemini Enterprise seats, region, admin controls)—align IdP/Cloud project and license assignment before rolling agents to managed users.
  • Approval policy: Codex --approve-for-me (0.147.0) plus safer cyber-model auto-review defaults (0.146.1)—document when auto-review is allowed vs human HITL; explain permission UX so reviewers trust the terminal, not screenshots.
  • MCP reliability: Windsurf v3.7.25 fixes auth with self-hosted GitLab and older Atlassian MCP—retest those connectors after client bumps; keep a short allowlist of MCP servers in shared runbooks.
  • Observability of agent cost/latency: Goose v1.46.0 per-message tokens/cost/TTFT/tok-s—useful for comparing loops in CI or dogfood channels without external scrapers.
  • SDK/host stop semantics: Cline sdk v0.0.72—queued turns surviving aborts and run.failed reporting matter for orchestration hosts; upgrade hosts before relying on “interrupt and continue” behavior.
  • Open improvement proposals (8) remain mostly team-audience P5 backports (older Codex/Cline/Goose/OpenCode/Zed/Grok)—triage separately from today’s adopt list; don’t confuse proposal age with this brief’s current versions.

What to ignore

  • Marketing tone and +N “more changes” bulk without behavioral impact—focus on flags, mode enforcement, auth, and data-loss paths called out above.
  • Prerelease noise: Gemini v0.56.0-preview.1 / nightly test-only bumps; Codex rust-v0.148.0-alpha.19 empty “Release alpha” signal.
  • Commit-synced monorepo blips (Grok be71313) until they land in a clear versioned release notes channel.
  • HTML-only changelog detail (Windsurf) when you need reproducible pins—pair with internal package hashes if you gate on exact bits.
  • Internal proposal links for older versions (e.g. Codex 0.144/0.145, Cline 4.0.7, Goose 1.44.0) as if they were today’s headline releases—they are backlog, not current evidence leaders.

Watch next

  • Whether Codex plugin catalogs and --approve-for-me spawn default team templates—or get restricted after cyber-model review defaults.
  • Cline plan-mode hard blocks spreading to other agents (policy-as-code vs prompt-only).
  • Claude Code identity forwarding adoption and any follow-on audit fields beyond spend headers.
  • Gemini CLI 0.56 leaving preview with caretaker/triage features that might affect issue-driven agent ops.
  • Goose cache-safe turn assembly + unrolled loop: stability under long MCP tool storms.
  • Antigravity enterprise controls interacting with sandbox/hooks/background surfaces tagged in-release.
  • Stable semver cut for Grok Build presence/worktree changes after the monorepo sync.

How to use DevAgentRadar

  • Read What the evidence shows as the only source of truth for versions—click through to the cited GitHub/changelog URLs before upgrading.
  • Developers: apply For developers as personal CLI/IDE checklist the same day.
  • Teams: turn For teams into policy tickets (approvals, MCP allowlists, identity/spend, enterprise sign-in)—don’t roll prereleases fleet-wide.
  • Use What to ignore in standup to kill rabbit holes; use Watch next for a light Friday scan.
  • Treat open proposals as optional backlog votes, not release evidence. This brief is judgment from curated signals, not vendor marketing.

Evidence appendix

  • Evidence window: top curated signals (versioned releases preferred)
  • Assistants tracked: 22
  • Signals in this brief: 16
  • Trusted source mix (GitHub/changelog vs HTML): 55%
  • Open improvement proposals: 8
  • Sources reporting errors: 0

High-signal releases used

  • Claude Code v2.1.233: Added GitLab merge request URL support to the --worktree flag and the claude agents view (where MRs display as !N) · Added an opt-in `forward_user_identit — source

  • OpenAI Codex CLI rust-v0.147.0: Install portable Agent Plugins and search across local, personal, workspace, and remote plugin catalogs. (#36544, #36409, #36919, #36796) · Organize conversatio — source

  • Zed AI v1.15.0: This week's release includes a new git.diff_base setting for choosing whether editor gutters, file-status colors, and git: diff show uncommitted changes aga — source

  • Google Antigravity 1.1.10: Added Business sign-in for Gemini Enterprise accounts, so you can authenticate with a Google Cloud project under Google Cloud terms, use a license seat allocate — source

  • Goose v1.46.0: Unrolled agent loop #9574 · Cache-safe request assembly with append-only turn context and declared cache semant — source

  • Cline desktop-v0.0.8: Edit any earlier message in a conversation — the app forks the session at that point, rewinds the workspace to that run's checkpoint, and re-runs from your edit — source

  • Gemini CLI v0.55.1: fix/verify release npm ci ignore scripts · fix(ci): prevent workspace binary shadowing in release verification · Feat/tool registry discovery · fix(ci): prevent — source

  • Cline v4.1.4: Recognize Chutes as a provider. · Show skills alongside workflows in the slash command menu, and disambiguate commands that share a name instead of letting one — source

  • Cline sdk/sdk/v0.0.72 — SDK v0.0.72: Prompts queued during a turn now survive being interrupted: they are preserved across user-initiated aborts, drained after a turn aborts itself, and edits made — source

  • Gemini CLI v0.56.0-preview.1: Changelog for v0.55.0-preview.1 · chore(release): bump version to 0.56.0-nightly.20260806.g761f604c1 · Changelog for v0.54.0 · Reclassifying Capacity Exhaustion — source

  • OpenAI Codex CLI rust-v0.146.1: Apply safer automatic-review defaults for cyber-capable models and explain permission changes in the terminal interface. (#37057) · #37057 [0.146] Backport safe — source

  • Gemini CLI v0.56.0-nightly.20260815.g2a87e7be1: [SSR Agent] Issue Fix (19826): Migrate process.env to vi.stubEnv in a2a-server tests — source

  • OpenCode v1.18.17: Made session compaction keep complete recent turns and produce clearer summaries for smaller models. · Added MERGE Gateway reasoning variants so those model opt — source

  • xAI Grok CLI (Grok Build) be71313 — Synced from monorepo: Synced from monorepo · Presence protocol end-to-end: live presence updates through the gateway to client presence tiers · Worktree: narrow standalone origin fet — source

Open proposals

Was this guidance useful?

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