A programming language for AI orchestrations
This repository also configures its own agents. See what weft tells them →
Skill Claude CodeCodex
Writes and overhauls the LLM prompts inside a weft program (system prompts in prompts/, user prompt templates, JSON extraction instructions). Runs on the WeaveMind prompt-building playbook. Dispatched by Tangle when a stage talks to a model and the prompt matters.
Skill Claude Code
Read when the program needs to answer calls from outside (an HTTP route, a websocket, anything a page or a service will call), and before briefing a frontend on what it will call: the URL known before activation, how a request's parts and body reach the graph, pictures in and out as links, streaming an answer as it is…
Skill Claude Code
Read before picking nodes for a job: the three reads that find a node and tell you how it wires, the metadata vocabulary you will meet there, the wiring shapes that recur (a connection, a provider, infrastructure, ports that come from a list), where a project's own nodes live, and what to do when nothing in the…
Skill Claude Code
Read when the user asks about connections, API keys, sign-ins or permissions: your part and the user's, the connect flow in the terminal and door by door in the editor, and what each banner means.
Skill Claude Code
Read when the user wants their own website, app, bot or extension to list, show, answer, skip or cancel what a program is waiting on, fetch a file it carries, or show what one of its nodes is displaying (a WhatsApp bridge's QR code, a database's minted password): the api token, the doors, the payload shapes, and how a…
Skill Claude Code
Read when telling the user where to click or what they are looking at in VS Code: the toolbar, the action bar, the palette, context menus, editing gestures, groups and loops in the graph, run and replay, the inspector, the connect flow, the problems panel. Labels are verbatim from the extension, except an infra node's…
Skill Claude Code
Read when the user wants a page, app or site for the program, and before dispatching the frontend-builder: the verified scaffold commands, the default stack (pnpm, SvelteKit, PostgreSQL, BetterAuth, shadcn-svelte), calling the program's own routes and signal doors, the three variables that say where the program is…
Skill Claude Code
Read when no node, group or node-smith contract can honestly deliver what the request needs, or when a node just built belongs in the shared catalog: the tracker to search first, the issue templates with their exact field ids, and the pre-filled issue URL to hand the user.
Skill Claude Code
Read before writing or editing any .weft source: declarations, wiring, config literals, reserved keys, inline signatures and expressions, types, groups and the level rule, included files, loops, the pulse execution model, @file, @asset and @include, and every compiler error slug.
Skill Claude Code
Read when the user asks what a run costs, or when a node is wired to a paid API weft does not already price: the three pieces of a custom paid service, a meter to copy, and the rules that keep a number honest.
Skill Claude Code
Read when a program calls a language model and you need it to behave: whether it reasons before answering and how hard, what bounds a reply's length and why one comes back empty, what a class of model costs in time and money, and how prompt caching works across a chain of calls.
Skill Claude Code
Read before dispatching a node-smith (the brief and the review) and when an expert writes a node by hand: the node authoring manual, the test tiers, and the shapes half-done work takes. The node-smith reads this same file as its manual.
Skill Claude Code
Read when someone asks to be taught, shown around or onboarded, asks what a thing on screen means, or opens a project they have clearly just made: the guided tour of the two views, the graph, building in it, running, the three things that are alive, executions, triggers, infra, connections, a public address and people…
Skill Claude Code
Read when running or activating a program: the CLI command map, the daemon (never yours to restart), running a command so it never asks and never sits quiet (with how long each normally takes), the build and run flow, trigger activation and the three deactivate modes, the infra lifecycle, and what each kind of trigger…
Skill Claude Code
Read when a program talks to a model, messages people, spends money or writes into other systems: the swiss cheese model, the free layers built by default, the layers that cost something and the one-sentence test for when a program is high stakes, the wiring shapes, and the red-teamer pass.
Skill Claude Code
Read when growing a program stage by stage: carving a run, supplying starting values, baking a trigger and firing it without activating, reusing completed work with a seed, freezing an accepted example, checking a new run against one, and what each refusal means.
Skill Claude Code
Read when the user asks to update or upgrade weft, or when something broke right after one: locating the checkout, the git pull plus setup.sh walk, what setup.sh rebuilds and what it preserves, and fixing a failed update.
Skill Claude CodeCodex
Builds the project's frontend under front/ end to end, unsupervised. Dispatched by Tangle with the pages a person must use and the signals behind them. Reads the weft-frontend and weft-consumers skills, builds on the default stack (pnpm, SvelteKit, PostgreSQL, BetterAuth, shadcn-svelte), holds the api token server…
Skill Claude CodeCodex
Builds exactly one weft node end to end, unsupervised. Dispatched by Tangle with a typed contract (one job, exact ports); researches the real API documentation when the node wraps a service, writes the node and extensive tests in the project's nodes/ folder, and proves it by running weft test-node on the local tiers…
Skill Claude CodeCodex
Attacks a weft program before it faces the world, hunting every credible hole: an outsider lying through an input, a hallucination the graph trusts, a rogue step with agency, an unguarded path to something that matters. Dispatched by Tangle when the program is high-stakes; reads the weft source, the prompts, and the…
Skill Claude CodeCodex
Deep post-mortem digging into a weft execution. Dispatched with an execution id or a symptom when the cause is unclear or the journals are long; reconstructs exactly what happened from events, logs, source, node code, and stored files, compares good runs against bad ones, and reports the finding with quoted evidence.…
Skill Claude CodeCodex
COMMAND, not reference: Validate and build the weft program, fix diagnostics until clean. Run this when the user asks for this step by name. The weft- reference skills beside it are things you read; this is a procedure you carry out.
Skill Claude CodeCodex
COMMAND, not reference: Inspect the last execution and chase the failure to one node. Run this when the user asks for this step by name, optionally naming an execution id. The weft- reference skills beside it are things you read; this is a procedure you carry out.
Skill Claude CodeCodex
COMMAND, not reference: Grow the program one stage at a time against a real input, seeded runs and frozen examples included. Run this when the user asks for this step by name, optionally naming the stage to grow next or an example to check. The weft- reference skills beside it are things you read; this is a procedure…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: