backend grok-cli (SuperGrok Heavy sub, $0 marginal) · window last 24h
Codex Security Review is in research preview: deeper PR security findings inline from repo context — wire this into your irreversible/security review tier as the CX seat’s automated PR reviewer, not a substitute for human gates. — x.com/gdb/status/2085496677725860064
Codex “Approve for Me” /
codex-auto-reviewcan eat ~95% of turns; fix = Projects or YOLO — if CX quota is vanishing, disable per-tool auto-review (Full access / yolo) or scope via Projects before you blame model choice. — x.com/steipete/status/2085581999927091567EMERGING: autonomy as a ladder — high autonomy / gated autonomy / human-as-must; quality = constraints + verification loops — rename your trust ledger in those terms (auto vs watch vs queue) so routing decisions match how top voices now talk. — x.com/addyosmani/status/2085618113199133067
Anthropic’s harness ban is only non–Claude Code (Pi, OpenCode, etc.); BYO Claude Code/Codex in apps like T3 still works; the tighter ban was delayed indefinitely — keep Claude Code as the official harness shell for Max billing; don’t panic about UI wrappers that invoke it. — x.com/theo/status/2085621311909642621
EMERGING: “Skillsets” — unlisted skill packs you can bundle/share for self, team, and agents — treat skill distribution as packs, not one-off SKILL.md copies; unlisted packs fit private fleet skills. — x.com/rauchg/status/2085779103169519853
GPT-5.6 Luna retested post ~80% price cut at same ARC scores; Free/Go get unlimited Luna, Plus/Pro get Sol + effort controls — reprice Luna into bulk/overflow routes and stop defaulting Sol/xhigh for mechanical work. — x.com/reach_vb/status/2085616395262599241
T3 Code shipped subagent + workflow observability (see what agents spawn) + 250 PRs of multi-thread harness work — even if you stay CLI-first, the durable pattern is observability of spawned subagents/workflows, not more planner prose. — x.com/theo/status/2085639979011891445
Grok Build: trust custom extensions on launch or helpers (e.g. Firstmate watchers) break; watchers should stay background and not block chat — one-time trust step on GK when using no-mistakes / Firstmate-style helpers. — x.com/kunchenguid/status/2085786670398578931
Highest-leverage action today: Audit CX Codex for codex-auto-review quota burn → Projects or yolo; enable Security Review on security/irreversible PRs.