16 stories
OpenAI just made the engine under Codex an Apache-2.0 platform — codex exec, the SDK, and the JSON-RPC app-server. The pitch is one number: harness changes alone took the same model from 13.3% to 38.3% on ARC-AGI-3 with 6x fewer tokens.
Meta's first terminal coding agent launched with a benchmark chart it loses — and a design worth stealing: background agents that persist all session, sub-agents fanned into isolated git worktrees, and a 12–21x discount if you let Meta train on your code.
Your personal setting is being ignored, or the repo's is — and the reason is that the direction of precedence is not the same for every mechanism. settings.json lets the repo beat your laptop; skills let your laptop beat the repo; CLAUDE.md lets nothing beat anything; permission rules refuse to…
The programmatic hooks you pass in options.hooks, not the shell hooks in settings.json — and the difference matters, because by default both fire in the same run. We registered a callback on all 31 events and ran it until we had the thing nobody publishes: the measured firing order.
The agent gets one shot, on a machine with no browser and nobody to ask — which changes which failures you get. We ran four headless arms on one real failing test: the locked-down run with no allowlist was denied Bash and Edit, wrote nothing, exited 0, reported subtype "success" and cost $0.383 —…
Seven questions a security review actually asks, answered from current documentation rather than a trust page — including the three where the honest answer is "nothing stops that".
The thing people reach for the moment they want to stop approving every command — and the one fear it does not remove. A real Dockerfile, a default-deny egress firewall measured before and after, and the four things that break first: a build-arg token that shows up in docker history three times…
The window is the constraint everything else follows from, and file reads are what fill it — about 31% of a representative session, against 1% for your own prompts. What /compact and /clear each do, and the part nobody warns you about: compaction does not bring everything back.
Google's terminal agent is the only one of the three with a free tier that works without a card — 1,000 model requests a day on a plain Google sign-in. The three auth paths and what each actually meters, why GEMINI.md silently ignores your AGENTS.md, the nested settings.json shape, MCP and…
The subscription-vs-API crossover is a computable number, not a matter of taste — and the tokens go almost nowhere people assume. We run a 40-turn Opus 5 session at $13.20 uncached and $2.86 warm, work the break-even for Pro, Max 5x and Max 20x, price the same session across four models at a clean…
Six extension mechanisms, and only three of them are different kinds of thing. What each one is, where it lives, when the model reaches for it, and the honest edge cases where two genuinely overlap — plus the two changes that broke everyone's mental model: custom commands are now skills, and MCP…
Everyone writes "put a SKILL.md in ~/.claude/skills". The real picture is two ecosystems with two installers that cannot see each other: marketplace plugins via claude plugin install, plain-git skills via npx skills.
Claude Code as a library — and the field almost everyone misreads. We ran the same agent twice: allowedTools is an auto-approve list, not a sandbox, and the first run called Bash despite Bash not being on it.
Precedence is not the same direction for every mechanism: settings.json lets the repo override your home directory, skills let your home directory override the repo, and CLAUDE.md files never override each other at all.
A complete stdio server in TypeScript, small enough to paste and run, then the parts that decide whether the model ever calls it: the tool description is prompt engineering rather than documentation, and an error should tell the model how to recover instead of just failing.
Install commands for macOS, Windows and npm, the OAuth vs API-key sign-in, the real Node floor, the telemetry-on-by-default caveat, and what a K3 session costs on a membership vs the API.
Showing 16 of 16 stories
You're all caught up.
Get the top AI news and tools straight to your inbox every week.
Weekly · AI news from Backgrind Blog
Join 15,000+ AI enthusiasts
