Installable Claude Code workflows for prompt caching, browser QA, subagents, bugfix loops, and team setup.
Two flags promise to stop Claude Code from pausing at every tool call. One of them reads your settings, honours your allowlist, and refuses to run anything g…
A skill that walks Claude Code through publishing your Claude stack to the Cables community in one conversation. No manual repo setup, no hand-written `stack…
Pick the stack that matches what you're building. Each one is a single `npx` command. Plugins installed, skills synced, marketplaces configured, no bash scri…
For building, debugging, and evaluating multi-step LLM agents. The skills that catch the problems traces-first: cache misses, regression in eval scores, trac…
The plugins and skills you want on day one of a typed Next.js app. From scaffold to prod, with CI/CD, observability, and release review baked in.
For designing, shipping, and validating product UI. Distinctive components without the AI-generic aesthetic, plus browser-QA to catch regressions before user…
*A five-part series tracing Anthropic's public thinking through Dario Amodei's writing and the company's model spec. One foundational document per entry, eac…
*Part 2 of 5. Tracing Anthropic's public thinking with FRE|Nxt Labs production commentary.*
*Part 3 of 5. Tracing Anthropic's public thinking with FRE|Nxt Labs production commentary.*
*Part 4 of 5. Tracing Anthropic's public thinking with FRE|Nxt Labs production commentary.*
*Final entry in the Amodei Playbook. From the safety floor of 2023 to the endgame framing of 2026.*
*A five-part series tracing Harrison Chase's public thinking on production AI agents. From LangChain as a weekend project to ambient agents. With FRE|Nxt Lab…
*Part 2 of 5. Tracing Chase's production agent thinking with FRE|Nxt Labs commentary.*
*Part 3 of 5. Tracing Chase's production agent thinking with FRE|Nxt Labs commentary.*
*Part 4 of 5. Tracing Chase's production agent thinking with FRE|Nxt Labs commentary.*
*Final entry in the Chase Playbook. From an 800-line weekend project to the UX that replaces chat.*
*A four-part series breaking down the Claude Code creator's actual workflow. Parallel agents, CLAUDE.md as postmortem, the "model wants tools" design philoso…
*Part 2 of 4. Breaking down the Claude Code creator's actual workflow with FRE|Nxt Labs production commentary.*
*Part 3 of 4. Breaking down the Claude Code creator's actual workflow with FRE|Nxt Labs production commentary.*
*Final entry in the Cherny Playbook. From origin story to design philosophy.*
If your E2E tests grew with the app, they almost certainly cover pages, not journeys. That distinction sounds academic until the most-broken flow in producti…
Every tool call Claude Code makes (a bash command, a file write, an MCP call) pauses and waits for approval. That is the right default. It is also death by a…
Claude Code can write code, run tests, stage files, and commit in one uninterrupted session. That is genuinely useful. It is also how you end up with `fix bu…
Passing unit tests are not a working product. Browser QA is the layer that asks the question your test suite can't: does the thing work for a human looking a…
Asking Claude Code to "fix this bug" without first reproducing it is the fastest way to close a ticket that reopens three days later. The fix will look plaus…
A `CLAUDE.md` that no longer matches the repo is worse than no `CLAUDE.md` at all. Stale context does not merely fail to help. It actively misdirects, and th…
The most common architectural mistake when building LangGraph agents is assuming the LLM can see your state fields. It cannot. The LLM only sees three things…
Building a production LangGraph agent requires middleware that injects dynamic content into the system prompt every turn. Prompt caching requires that the sy…
When you add sub-agents to a production LangGraph system, the instinct is to build them as a separate agent type with its own state schema, its own graph str…
*A six-part series tracing Andrej Karpathy's public thinking from 2017 to 2025. One foundational post per entry, each with FRE|Nxt Labs live commentary on ho…
*Part 2 of 6. From Software 2.0 to the March of Nines, tracing Karpathy's public thinking with FRE|Nxt Labs production commentary.*
*Part 3 of 6. Tracing Karpathy's public thinking with FRE|Nxt Labs production commentary.*
*Part 4 of 6. Tracing Karpathy's public thinking with FRE|Nxt Labs production commentary.*
*Part 5 of 6. Tracing Karpathy's public thinking with FRE|Nxt Labs production commentary.*
*Final entry in the Karpathy Playbook. From 2017's Software 2.0 to 2025's reality check.*
Publishing cables is not just writing markdown. The failures usually happen in validation, index generation, or submodule pointer updates.
Manual QA catches issues, but it does not scale when UI changes land every day.
Without a `CLAUDE.md`, every session starts from zero. The model has to re-derive your conventions, package manager, folder layout, and definition of done on…
The rule is simple and almost everyone discovers it the hard way: `~/.claude/` is for patterns you want everywhere, and `.claude/` in the repo is for pattern…
A skill is a reusable operating procedure written in markdown. The value is consistency, not cleverness. The first skill you write should replace the prompt…
A subagent is not a clever trick for parallelism. It is a way to keep the parent's context clean while a narrowly-scoped task runs in its own. The right use…
A slash command is the fastest way to package a routine you run every week. It is not a place for reasoning; it is a place for the steps you already trust, n…
One `CLAUDE.md` works until the project's complexity exceeds a single file's clarity budget. The symptom is not size; it is that edits in one section start b…
Most plugin bloat comes from installing tools before defining workflows. A crowded command palette is not capability. It is a tax on every new teammate who h…
Cables earn trust by describing what actually happened, including the part where things did not work. A war story is not a polished retrospective; it is the…
We had tests. We also had regressions. The problem was coverage shape, not raw test count.
The hardest part of a team rollout was not the tooling. It was the quiet drift toward six slightly different workflows sharing one repo.
Subagents did not speed us up until we redesigned the boundaries of the work itself. Parallelism without ownership does not compound; it collides. The levera…
Once more than one person relies on `CLAUDE.md`, it stops being notes and starts being a contract. Treat it as personal scratchpad and you get three engineer…
A skill that works only for its author is not a skill; it is a shell alias with extra steps. Most skills fail at team scale because they assume context the a…
You cannot optimize what you do not measure. For agent systems, that means evaluating both quality and execution cost. And doing it per task class, not in ag…
When an agent fails, raw logs usually tell us *what* happened, not *why* it happened in that step sequence.
Prompt caching is one of those optimizations we all think we enabled "well enough" until costs spike and response times drift.
The fastest way to ship bugs is to skip one of those three steps.
Claude Code is not a chat window with a terminal attached. It is two products fused at the prompt: a conversation, and a command palette. Until you find the…