Radar brief — 2026-08-06
Security hardening and session control dominate: Claude Code masks sandbox credentials, Codex tightens cyber-model review, Cline locks plan mode, and Goose ships a critical CLI advisory—while agent plugins and enterprise auth keep expanding.
What the evidence shows
Ctrl+Alt+F); Linux/WSL sandbox gains mode: "mask" for credential files (macOS falls back to deny)—direct agent UX + secret-handling upgrade/import migrates Cursor/Claude settings, MCP, plugins, sessions; experimental Bedrock login with GPT-5.6 Sol defaultanthropic/claude-sonnet-5goose review / git core.fsmonitor; plus latest Gemini models, Stop-hook working_dir, Sakana/Fugu providerGoogleCredentialsAuthProvider (cleartext leak fix); rotate session ID on model fallback to avoid stateful API errorsgit.diff_base chooses HEAD vs default-branch merge base for gutters/diff; Project Panel → external app drag (macOS/Linux Wayland); JSX/TSX linked editing + Emmet in arrow/return bodiesTheme read: model / agent / cli / mcp / security cluster tightly. Breaking or behavior-changing surfaces show up in Claude sandbox masking, Codex session/plugin model, Cline plan-mode enforcement, and Goose’s advisory.
Source mix: ~55% GitHub/changelog-style. Windsurf is HTML changelog; Grok is a monorepo sync commit (shorter SHA, noisier than tagged releases). Gemini nightlies/previews overlap stable notes—treat as in-flight.
For developers
- Upgrade Goose to v1.44.0 now if you use
goose reviewor shared git config—GHSA-r5pp-p5r8-466r is arbitrary command execution viacore.fsmonitor. - Claude Code v2.1.221: try Focus view when long tool traces drown the chat; on Linux/WSL, prefer sandbox
mode: "mask"for credential files instead of wholesale deny where you still need egress substitution. - Cline v4.1.4: plan mode is stricter—expect refused
sed/git commit/package installs/redirection; drive plan→act yourself. Use the slash menu for skills vs workflows when names collide. - Cline desktop-v0.0.8: use mid-thread edit + checkpoint fork to rewind bad turns without half-applied workspace state.
- Codex CLI 0.145→0.146.x:
/importfrom Cursor/Claude;/new//clear, pin, and fork threads; if you touch cyber-capable models, take 0.146.1’s safer auto-review defaults. - Gemini CLI v0.54.0: HTTPS credential path fix matters on any non-trivial network path; session-ID rotate on model fallback reduces sticky API failures.
- OpenCode v1.18.5: worth it if you hit Claude thinking-shape breaks, Mistral cache drift, or MiniMax M3 variant mis-selection.
- Grok Build ed6d543: MCP image extraction before truncation and subagent cap (16) change long-run behavior—retest workflows that stream images or fan out agents.
- Windsurf v3.6.27: required behind corporate TLS inspection on Windows; symlink write refusal changes how “approved edit” paths behave—don’t rely on symlink retargeting.
For teams
- Secret egress policy: Claude’s mask-mode sandbox and Windsurf’s symlink write block are concrete controls—document them in agent allowlists alongside MCP server inventories.
- Plan/act governance: Cline’s hard block on mutating shell in plan mode is a process win—align team norms so models cannot self-escalate to edits.
- Codex plugin + import surface: Agent Plugins manifests, extra marketplaces (Bedrock, Claude Code), and Cursor/Claude migration expand supply chain—review which marketplaces are trusted before org-wide rollout.
- Enterprise auth: Google Antigravity 1.1.10 Business sign-in ties seats, region, and admin controls to Gemini Enterprise—pilot under a dedicated Cloud project before defaulting individuals to consumer auth.
- Codex 0.146.1 cyber-model defaults: if security/research agents run in shared orgs, standardize on the safer automatic-review defaults and surface permission diffs in PR templates.
- Goose advisory: treat as a fleet patch; audit developer machines for unexpected
core.fsmonitorgit config after exploit disclosure. - Zed v1.15.0-pre:
git.diff_basechanges what “dirty” means in review UX—agree team-wide (HEAD vs merge-base) before mandating the pre-release.
What to ignore
- OpenCode
pr-38252-videos: verification recordings only—no product signal. - Gemini CLI nightlies/previews (
v0.54.0-preview.0,v0.55.0-nightly.*) whenv0.54.0already carries the actionable HTTPS and session fixes—don’t chase duplicate changelog noise. - Raw “+N more changes” volume on Codex/Claude/Zed as a quality proxy—prioritize security, plan-mode, and session/plugin surfaces above change counts.
- Open proposals at P5 (Codex 0.145/0.144.6, Cline v4.0.7, Grok 8adf901, OpenCode videos/v1.18.5, Zed v1.13.0-pre, Goose v1.44.0)—backlog hygiene, not urgent radar movers unless they match a security path you’re already on.
- Marketing framing around model names (e.g., Bedrock default model labels)—validate in your account; evidence here is release-note level only.
Watch next
- Codex plugin marketplace adoption and whether Bedrock/Claude Code manifests stabilize beyond experimental flags.
- Claude Code credential mask behavior on macOS (still deny-fallback) and any follow-up for parity.
- Goose post-GHSA residual risk: related git-hook/fsmonitor patterns in other CLIs.
- Cline plan-mode hard blocks vs real repo workflows (monorepo installs, codegen scripts)—expect community pushback or finer policy knobs.
- Antigravity Enterprise admin control coverage: which agent features actually inherit org policy.
- Grok ACP session resume/close and subagent cap under heavier MCP image workloads.
- Zed v1.15.0 stable after the
-pregit.diff_base and drag-out changes.
How to use DevAgentRadar
- Read What the evidence shows first—versioned GitHub/changelog rows only; no invented releases.
- Developers: patch security and behavior breaks (Goose advisory, Cline plan mode, cert/symlink fixes), then adopt session/UX wins (Focus view, Codex forks, checkpoint edit).
- Teams: convert those into policy—trusted plugin marketplaces, plan/act escalation rules, enterprise auth pilots, fleet advisory SLAs.
- Skip What to ignore noise; track Watch next for unfinished surfaces (plugins, mask parity, enterprise controls).
- Evidence window is curated signals across 22 tracked assistants (17 in this brief); treat HTML changelogs and monorepo sync commits as lower confidence than tagged GitHub releases.
Evidence appendix
- Evidence window: top curated signals (versioned releases preferred)
- Assistants tracked: 22
- Signals in this brief: 17
- Trusted source mix (GitHub/changelog vs HTML): 55%
- Open improvement proposals: 8
- Sources reporting errors: 0
High-signal releases used
-
Claude Code
v2.1.221: [VSCode] · Added Focus view: a chat-menu toggle that hides tool activity behind an expandable per-turn summary with a live running-tool indicator, toggled with — 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 -
Zed AI
v1.15.0-pre: This week's release includes a newgit.diff_basesetting for choosing whether editor gutters, file-status colors, andgit: diffshow uncommitted changes aga — source -
OpenAI Codex CLI
rust-v0.145.0: Added experimental paginated thread history with efficient resume, search, persisted names, sub-agent support, and memories. (#33364, #33907, #34085, #34229, #3 — 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 -
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 -
Goose
v1.44.0: Arbitrary command execution in goose CLI viagoose review(via git core.fsmonitor config) - [GHSA-r5pp-p5r8-466r](https://github.com/aaif-goose/goose/security — 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 -
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
v0.54.0: Changelog for v0.53.0-preview.0 · Changelog for v0.52.0 · chore(release): bump version to 0.54.0-nightly.20260722.gf743ab579 · fix(caretaker): sanitize and wrap — source -
OpenCode
v1.18.5: Improve Claude adaptive thinking handling across more response shapes. · Avoid OpenAI Responses phase handling that could break some conversations. · Preserve g — source -
Gemini CLI
v0.54.0-preview.0: Changelog for v0.53.0-preview.0 · Changelog for v0.52.0 · chore(release): bump version to 0.54.0-nightly.20260722.gf743ab579 · fix(caretaker): sanitize and wrap — source -
Gemini CLI
v0.55.0-nightly.20260729.g3499c84f7: chore/release: bump version to 0.54.0-nightly.20260728.gbef611950 · feat(pr-generator-db): implement Firestore concurrency dual-locking and test ingestion utili — 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.