Radar brief — 2026-08-18

CLI agents double down on approvals, plan-mode guardrails, and enterprise auth, while plugin catalogs and session-fork workflows keep shipping—security and control surfaces matter as much as model upgrades this window.

What the evidence shows

Versioned GitHub-style releases dominate the top scores; one HTML changelog and several prereleases/nightlies are noisier.

Claude Codev2.1.234

Key Signal & Impact

Optional CLAUDE_CODE_PROJECT_DIR_NAME for per-session config hosts; selection:clear keybinding (incl. agents view); GitLab MR badge via authenticated glab (!N draft/pending/green). Large multi-surface drop (+48).

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 without dupes (+155).

OpenAI Codex CLIrust-v0.146.1

Key Signal & Impact

Safer automatic-review defaults for cyber-capable models; permission changes explained in the TUI—direct security posture tweak.

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 (macOS/Linux Wayland); JSX/TSX linked editing + Emmet in arrow/return bodies.

Goosev1.46.0

Key Signal & Impact

Unrolled agent loop; cache-safe append-only turn context; streaming shell output; per-message tokens/cost/TTFT/tok/s + session totals.

Google Antigravity1.1.10

Key Signal & Impact

Business sign-in for Gemini Enterprise (Cloud project, license seats, regional inference, org admin controls).

Gemini CLIv0.55.1

Key Signal & Impact

Security: case-insensitive sensitive-path blocklist + VS Code HITL; defensive @ path resolution; tool-registry discovery; release-pipeline hardening.

Clinev4.1.4

Key Signal & Impact

Plan mode hardened: user-driven exit only; hard-block file-editing shell (redirects, in-place editors, mutating git, installs); skills next to workflows in slash menu; Chutes provider.

Clinedesktop-v0.0.8

Key Signal & Impact

Edit any earlier message → fork + workspace checkpoint rewind; transactional/atomic restores; more reliable checkpoints after restart/compaction/resume.

Clinesdk/sdk/v0.0.72

Key Signal & Impact

Queued prompts survive aborts; durable session context across hub restarts; failed queued turns report run.failed.

OpenCodev1.18.17

Key Signal & Impact

Compaction keeps complete recent turns; capped retries + jitter; PDF for Copilot models with vision; provider/model routing fixes (DeepSeek V4 Flash, Muse, MERGE Gateway).

Windsurfv3.7.25

Key Signal & Impact

Devin Desktop sidebar perf for huge session caches; MCP auth fixes (self-hosted GitLab, older Atlassian). HTML changelog—treat as secondary to GitHub tags.

xAI Grok CLI (Grok Build)be71313

Key Signal & Impact

Monorepo sync: presence protocol, worktree fetch hygiene, pager/tool-card UX, update telemetry. Commit-level signal, not a semver tag.

Lower-confidence / thin signals: Gemini CLI v0.56.0-preview.1 and nightly v0.56.0-nightly.20260818… (caretaker/SSR/TUI fixes, prerelease); Codex rust-v0.148.0-alpha.22 (tag only, no substantive notes in evidence).

For developers

Codex (rust-v0.147.0): Try Agent Plugins and catalog search locally; use conversation sections for long threads. Prefer rust-v0.146.1+ if you run cyber-capable models so auto-review defaults are the safer set. Inspect --approve-for-me before relying on it for unattended runs.
Cline (v4.1.4): Expect plan mode to refuse mutating shell/file ops—not just prompt against them. Drive plan→act yourself. Use slash menu for skills vs workflows when names collide.
Cline Desktop (desktop-v0.0.8): Use mid-thread edit + checkpoint rewind to experiment without manual git reset; trust failed restores less—restores are intended to be atomic.
Goose (v1.46.0): Lean on streaming shell + per-message cost/TTFT UI to debug slow or expensive turns; cache-safe assembly is relevant if you care about prompt-cache hit rates.
Claude Code (v2.1.234): Set CLAUDE_CODE_PROJECT_DIR_NAME if your host isolates config dirs; bind selection:clear; wire glab auth to get MR state in the footer.
Gemini CLI (v0.55.1): Pull for path-blocklist and @-reference resolution fixes before filing path-escape bugs.
Zed (v1.15.0): Set git.diff_base explicitly so gutters match whether you think in “uncommitted” vs “branch vs main.”
OpenCode (v1.18.17): Useful if compaction or retry storms have been noisy; check PDF attach only on Copilot models that advertise vision.

For teams

Enterprise auth path: Antigravity 1.1.10 Business sign-in (Gemini Enterprise seats, region, admin controls) is the clearest org-readiness signal this window—pilot under a dedicated Cloud project before broad rollout.
Approval policy: Codex’s --approve-for-me plus safer cyber-model auto-review (0.146.1 → 0.147.0) needs a written policy: who may enable auto-approve, on which repos, with what audit.
Plan-mode as control surface: Cline’s hard blocks in plan mode are enforceable guardrails—good default for supervised agents; document that models can no longer self-switch to act.
Session integrity: Cline Desktop transactional rewinds + SDK queue/abort durability reduce “half-applied” workspace risk—prefer these builds for shared agent hosts.
MCP estate: Windsurf’s MCP auth fixes (GitLab/Atlassian) and broad surface:mcp tags across Claude/Codex/Gemini/Cline justify a quick matrix test of self-hosted MCP against your pinned client versions.
Open proposals (P5 backlog): older improve/review items still open for Codex 0.145.0/0.144.6, Cline v4.0.7, Goose v1.44.0, OpenCode v1.18.5, Zed v1.13.0-pre breaking review, Grok sync—triage against current tags above before adopting stale proposal targets.

What to ignore

Codex rust-v0.148.0-alpha.22 — empty release note in evidence; not actionable.
Gemini nightlies/previews (v0.56.0-*) — internal caretaker/triage and SSR fix chatter; wait for stable notes unless you dogfood previews.
Marketing-shaped HTML blurbs without versioned GitHub detail (Windsurf is usable but secondary).
Raw “+N more changes” volume — score and concrete control/security bullets beat changelog length.
Stale open proposals pinned to older versions when newer evidence (e.g. Codex 0.147, Cline 4.1.4, Goose 1.46) already supersedes them.

Watch next

Whether Codex plugin catalogs and --approve-for-me pick up tighter default deny lists after the cyber-model review change.
Cline plan-mode hard blocks: any escape hatches or team policy docs in follow-on tags.
Gemini CLI stable after the 0.55.1 security path fixes—does 0.56 ship tool-registry discovery cleanly?
Goose production impact of unrolled loop + cache-safe assembly on cost and latency.
Antigravity Enterprise: admin-control coverage beyond sign-in (sandbox/hooks tags already present).
Claude Code follow-through on tagged surfaces (permissions, context-files, background, MCP) called out in release tags but not fully expanded in the summary text.
Grok Build: first proper semver/tag after monorepo commit be71313.

How to use DevAgentRadar

Treat this as an evidence radar, not a vendor ranking—act only on versions and behaviors listed above.
Developers: pin one CLI/agent bump this week that matches your pain (approvals, plan guards, checkpoints, or cost UI) and verify on a throwaway repo.
Teams: map approval, plan-mode, MCP auth, and enterprise sign-in signals to policy before wide enablement; close or retarget open proposals to current tags.
Re-check source URLs (GitHub releases first) when you adopt; ignore alphas/nightlies unless you explicitly dogfood.
Suggest improvements via open proposals when a release lacks security/rollback notes your org needs.

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 v2.1.234: Added the optional CLAUDE_CODE_PROJECT_DIR_NAME environment variable: hosts that give each session its own config directory can choose a short name for the pe — 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
Goose v1.46.0: Unrolled agent loop #9574 · Cache-safe request assembly with append-only turn context and declared cache semant — 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
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 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
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.20260818.g194edea47: [SSR Agent] Issue Fix (26120): Clarify privacy notice wording and selection options · [SSR Agent] Issue Fix (21919): Fix TypeScript strict-null errors in integr — 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
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