Radar brief — 2026-08-12
Agent runtimes tighten control surfaces: plan-mode hard blocks, safer auto-approve defaults, self-hosted runners, and portable plugins dominate this window’s versioned releases.
What the evidence shows
High-signal, versioned GitHub releases cluster on agent loop control, plugin/skill portability, CLI approval/sandbox posture, and enterprise auth. MCP and model tags are dense across the set; several notes are explicitly breaking or security-tagged.
git.diff_base chooses gutter/diff baseline (HEAD vs default-branch merge base); Project Panel drag-out on macOS/Linux Wayland; linked editing + Emmet in JSX/TSX bodiesclaude self-hosted-runner for Team/Enterprise web/mobile/desktop sessions on your machines; archive plugin source (HTTPS zip + optional SHA-256 pin); Bedrock ANTHROPIC_BEDROCK_REGION_PREFIX--approve-for-me auto-reviewed approvals; import/sync Cursor & Claude skills/conversations without dupes/new//clear session naming, pin/switch side threads; plugin manifests + Bedrock/Claude Code marketplaces; thread fork (incl. temporary); app-server ↔ remote Code Mode over WebSocketanthropic/claude-sonnet-5Lower-weight / noisier in this mix: Gemini CLI v0.56.0-preview.1 and nightly (prerelease caretaker/triage work), xAI Grok CLI commit be71313 (monorepo sync, not a clean semver tag), Windsurf v3.7.16 (HTML changelog; Devin Desktop restart confirmation, settings-as-tabs). One tracked source reported errors; treat incomplete changelogs cautiously.
For developers
- Cline v4.1.4: Stop relying on the model to leave plan mode—switch yourself. Expect shell file edits, redirects, mutating git, and package installs to be refused in plan mode; use read-only investigation there, act mode for writes.
- Cline desktop-v0.0.8: Use mid-thread message edit to fork + checkpoint-rewind when a prompt went wrong; failed restore should not half-apply—still verify workspace state after long runs.
- Codex CLI rust-v0.147.0 / 0.146.x: Try portable Agent Plugins and skill import from Cursor/Claude; if you enable
--approve-for-me, read the rust-v0.146.1 cyber-model auto-review default changes before unattended runs. - Claude Code v2.1.224: Pin plugins via
archive+ SHA-256 when you cannot use git/npm; setANTHROPIC_BEDROCK_REGION_PREFIXif Bedrock cross-region profile selection matters. - Goose v1.46.0: Use streaming shell + per-message cost/TTFT stats to debug slow/expensive turns; cache-safe turn assembly is relevant if you care about prompt-cache hit stability.
- Zed AI v1.15.0: Set
git.diff_basedeliberately so gutters/git: diffmatch your review habit (uncommitted vs branch-vs-default merge base). - Gemini CLI v0.55.1: Update for path blocklist / at-reference resolution fixes before trusting
@file refs on shared machines. - OpenCode v1.18.10: Modal model auto-discovery reduces manual model list churn; minor UI stability fixes only.
For teams
- Enterprise pathing: Antigravity 1.1.10 Business sign-in ties Gemini Enterprise seats, region, and admin controls to the CLI—align IdP/Cloud project and license assignment before rollout.
- Claude Code self-hosted runners (Team/Enterprise): Treat
claude self-hosted-runneras a capacity and data-residency control plane; define which machines/containers may host web/mobile/desktop sessions. - Approval policy: Codex
--approve-for-meplus safer cyber-model defaults (rust-v0.146.1) needs a written policy: which repos, which models, mandatory human review for privileged paths. - Plan-mode integrity (Cline v4.1.4): Hard blocks beat prompt-only guardrails—good baseline for shared agent norms; document plan vs act for contributors.
- Plugin supply chain: Claude
archiveHTTPS zip + SHA-256 pin and Codex plugin catalogs/marketplaces (incl. Bedrock/Claude Code) need an allowlist and hash review, not ad-hoc install. - Cost observability: Goose per-message usage/cost UI supports chargeback and regression checks on agent loops—standardize on recording session totals in team retros.
- Open proposals (8) still point at older Codex/Cline/Goose/OpenCode/Zed/Grok revisions; prefer the newer versions in this brief when closing those P5 items.
What to ignore
- Marketing tone or undated blog summaries—not in this evidence set; stick to the tagged GitHub releases above.
- Gemini
v0.56.0-preview.1/ nightly caretaker stacks until they land in a stable tag (prerelease noise). - xAI Grok CLI
be71313monorepo sync as a version pin—commit-level, mixed bag; wait for a clearer release boundary. - Windsurf HTML changelog items that are IDE chrome renames (“Plugins” → “Extensions”) unless you operate Devin Desktop multi-agent restarts.
- Open improvement proposals aimed at older versions (e.g. Codex 0.144/0.145, Cline v4.0.7, Goose v1.44.0) when this window already has newer tags.
- Internal scoring fields or “+N more changes” bulk without reading the named security/breaking bullets.
Watch next
- Whether Codex plugin manifests +
--approve-for-mepick up more marketplace/default-policy tightening after rust-v0.146.1’s cyber-model defaults. - Claude Code self-hosted runner adoption constraints (plans, sandbox/permissions surfaces already tagged on v2.1.224).
- Cline plan-mode enforcement follow-through in SDK (
sdk/v0.0.67reasoning normalization) and desktop checkpoint fork UX. - Gemini CLI stable after v0.55.1 security path fixes—does caretaker work promote cleanly out of preview/nightly?
- Goose unrolled loop + cache-safe assembly: measurable latency/cost wins in real MCP/tool-heavy sessions.
- Zed
git.diff_baseas a team default for agent-generated diffs in monorepos.
How to use DevAgentRadar
- Read What the evidence shows first—only versioned facts from this window’s JSON, not vendor hype.
- Developers: apply the concrete CLI/IDE actions on your machine the same day (flags, plan mode, plugin pins).
- Teams: turn approval, self-hosting, and plan-mode blocks into written defaults; close stale open proposals against current tags.
- Discount prerelease/nightly/commit-only and HTML-only notes until they reappear as stable GitHub releases.
- Re-scan next brief for follow-through on security defaults, plugin catalogs, and enterprise auth—not for feature checklist inflation.
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: 1
High-signal releases used
-
Zed AI
v1.15.0: This week's release includes a newgit.diff_basesetting for choosing whether editor gutters, file-status colors, andgit: diffshow uncommitted changes aga — source -
Claude Code
v2.1.224: Added self-hosted environments:claude self-hosted-runnerturns your own machines or containers into a place Claude Code web, mobile, and desktop sessions can — 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 -
Goose
v1.46.0: Unrolled agent loop #9574 · Cache-safe request assembly with append-only turn context and declared cache semant — source -
OpenAI Codex CLI
rust-v0.146.0: Name new sessions with/newor/clear, pin important threads, and switch between side conversations without closing them. (#34605, #34840, #35011) · Support — 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/v0.0.67— SDK v0.0.67: Reasoning controls (effort, budget, on/off) are now driven by the models.dev catalog and normalized once before provider encoding, so requests match what each p — 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.20260808.gcf22ac7e8: Reclassifying Capacity Exhaustion as Terminal Error · feat(caretaker): update Firestore schema with error, and pr_number fields · feat(caretaker-triage): prompt — source -
OpenCode
v1.18.10: Discover available Modal models automatically. (@devennavani) · Prevent adding the same attachment more than once. · Always show the new session button. · Impro — source
Open proposals
-
Improve with OpenAI Codex CLI rust-v0.145.0: rust-v0.145.0 0.145.0 (audience: team, P5)
-
Improve with Cline v4.0.7: v4.0.7 (audience: team, P5)
-
Improve with xAI Grok CLI (Grok Build) 8adf901: Synced from monorepo (audience: team, P5)
-
Improve with OpenCode pr-38252-videos: pr-38252-videos PR #38252 verification videos (audience: everyone, P5)
-
Improve with OpenAI Codex CLI rust-v0.144.6: rust-v0.144.6 0.144.6 (audience: developer, P5)
-
Review breaking changes in Zed AI v1.13.0-pre (audience: team, P5)
-
Improve with OpenCode v1.18.5: v1.18.5 (audience: team, P5)
-
Improve with Goose v1.44.0: v1.44.0 (audience: team, P5)
Anonymous · one vote per visitor · re-click to clear. Helps improve digests—not a product ranking.