Blog
Notes from the registry
Guides, deep dives, and ecosystem notes on agent skills — how they work, how to install them in Claude Code, Codex, Copilot, and Antigravity, and how to get the most out of your AI coding agent.

Best models for agent skills in October 2026: a routing table by skill type
Which model should run which agent skill step in October 2026: a routing table by skill type, Artificial Analysis v5 scores and a worked per-run budget.
Read post
What Skill Authors Can Learn from Caveman's SKILL.md
A skill-author review of Caveman v4.0: terse-output rules, activation, progressive disclosure, composing with other skills, and how to measure your own SKILL.md.
Read post
Supervising Claude Code skill and subagent runs from your phone
Supervise long Claude Code skill and subagent runs from your phone with Remote Control: which prompts reach you, what can run unattended, which rules to set.
Read post
Copilot AI Credits for agent skills: budgeting skills, subagents and custom agents
Price SKILL.md runs, subagent fan-out and custom agents under GitHub Copilot AI Credits: worked costs, the 30-credit CLI floor, budgets and Fusion.
Read post
Running agent skills through Desktop Commander: a pre-flight guide for skill authors
What changes when skill scripts run through Desktop Commander MCP: write limits, directory scope, vetting, containers, remote jobs and a checklist.
Read post
Running agent skills and subagents on GLM-5.3-Flash: cost per run and where it breaks
What a skill run or subagent fan-out costs on GLM-5.3-Flash at Z.ai list prices, the tool-calling limits to plan for, and when to route skills to GLM-5.3.
Read post
Kimi K3.1 Pro for agent skills: setup, cost per run, and rate limits
Run SKILL.md workflows and subagents on Kimi K3.1 Pro: harness setup, worked cost per skill run, cache effects, API tiers, and subscription quota.
Read post
Codex usage limits for agent workflows: skills, subagents and long runs
How skills, subagents, long sessions and model choice drain Codex's 5-hour and weekly pools, and how to budget a skill-heavy workflow. Verified Oct 28, 2026.
Read post
Routing agent skills across GPT-6 Astra, Sol and Luna
Which GPT-6 tier should run which agent skill step: prices, cached SKILL.md context, Codex usage and a worked per-run budget.
Read post
Opus 5.5 vs Sonnet 5.5 for agent skills: which model and effort per skill
Which Claude model and effort level to run skills and subagents on in Claude Code: Opus 5.5 vs Sonnet 5.5 routing, SKILL.md token cost and fan-out math.
Read post
Muse Wonder, scored row by row: what Meta's frontier model does and doesn't prove
Meta's Muse Wonder tops 7 of 14 rows on its own scorecard and trails Claude Opus 5.5 on 6. A row-by-row audit of benchmarks, price, access and the model card.
Read post
Tavus Griffin: the first model to pass the video Turing test
Tavus Griffin is the first Human Interaction Model and first to pass a video Turing test: 48% of live callers thought it human, and it tops NVIDIA VideoFDB.
Read post
Gemini 4 Argon: Google's frontier model, explained
Google DeepMind's Gemini 4 Argon: the 1M-token output limit, the vendor-reported benchmarks, Fairwind access, introductory pricing, and what developers can plan for.
Read post
Codex Security Cloud: the four-stage pipeline and its limits
How Codex Security Cloud scans GitHub repos: the threat model, validation sandbox, remediation flow, and the limits OpenAI states itself.
Read post
GPT-6.1 Sol for coding agents: benchmarks, pricing, and fit
GPT-6.1 Sol pricing, what the benchmark configurations actually say, where cached input changes the maths, and which agent workloads fit.
Read post
OpenAI DevDay 2026: the 25 announcements, mapped for agent builders
All 25 OpenAI DevDay 2026 announcements from the official recap, with availability status and what each changes for an agent build.
Read post
OpenAI dots: the permission model and what to hand off
How OpenAI dots work: Auto-review, Custom Rules, read-only background research, the steps a dot hands back, and what to delegate.
Read post
Ultrafast and the Pro tiers: a cost-planning guide
What Ultrafast's 8x speed costs in included usage, how Pro 100, 200 and 500 differ, and the grandfathering date to check before switching.
Read post
TypeSafe Jev: the model that answers instead of writing
Jev, the first System One model from ChatGPT co-inventor Diogo Almeida's TypeSafe: $0.042/MTok input, free output tokens, 70-500ms decisions, no hallucinated types. How it works, where it fits, and what to distrust.
Read post
Gemini 3.8 Live & Extended Thinking: the developer's guide
Google's two new real-time voice models on the Live API: session config changes, the Extended Thinking interaction_status protocol, pricing, and the migration path from 3.1 Flash Live.
Read post
We Must Pace the Frontier: Inside Amodei's Three-Step Plan to Slow AI
Amodei's essay asks the AI industry to slow down. The verbatim three-step plan, what Anthropic committed to, who signed on, and the pushback.
Read post
What Are Agent Skills? The Complete Guide
Agent skills are reusable instruction packs that teach AI coding agents new workflows. How they work, how to install them in Claude Code, Codex, Copilot, and Antigravity, and how to pick good ones.
Read post