A single skill
One stave, one job. It's a folder, not just a file: a README, an AGENTS.md or SKILLS.md, and whatever else the task needs. Staves are versioned and forkable, so the good ones are easy to build on.
Browse staves →About Glyph
An agent is only as good as the instructions behind it. Glyph is the open registry where people write those instructions down, version them, and share them — for Claude, Codex, a local model, or anything that reads instructions. One skill is a stave. A stack of staves is a weave. That's the whole idea.
Work in progress. These ideas are still settling — expect the model and its best practices to change.
The hard part of working with an agent isn't the model — it's telling it how you want the work done. People solve that once and then lose it: the prompt that finally nailed a code review, the workflow that cleaned up messy data, buried in a notes app or a private repo. The next person starts from zero. Glyph is the shared place to put that work, so a good stave gets found and built on instead of rewritten.
The whole platform rests on a small pattern: a stave is one skill, a weave is a stack of them that runs on its own, and a personality is an optional voice you can pull over the top. Learn those three and you know how Glyph works.
One stave, one job. It's a folder, not just a file: a README, an AGENTS.md or SKILLS.md, and whatever else the task needs. Staves are versioned and forkable, so the good ones are easy to build on.
Browse staves →A weave puts staves in a deliberate order. It tracks each stave by family, not by snapshot, so it can follow the latest version or pin a known-good one when you need the output to stay put.
Browse weaves →A personality is a stave of a different kind: it doesn't do the work, it sets how the agent shows up — its voice, how it opens, and an optional intent that colors every reply. Pull one over a weave and the whole run takes on that character.
Browse personalities →A stave is just markdown and files — nothing proprietary, nothing that only runs inside Glyph. Any agent that reads instructions can use one, so what you install here works the same whether you point it at Claude, Codex, or a model running on your own machine.
# Stave: Code Reviewer ## Role You are a meticulous reviewer focused on security and performance. ## Instructions 1. Flag SQL injection vectors 2. Check for N+1 queries 3. Verify error handling
Found a stave you want? There are two ways to put it to work. It's plain markdown either way — nothing has to run inside Glyph.
Add /mdto any stave's URL and paste it into a browse-capable agent. It fetches the stave as clean markdown and adopts it on the spot — no download, no setup.
…/staves/code-reviewer/md
Or download the stave and drop the folder straight into your agent — Claude, Codex, or a model on your own machine. Same files, run wherever you want.
A weave already runs on its own, so there's nothing to wire up. A personality is the optional layer that sits on top: it doesn't change what the work is, it changes who's doing it. Build one, cap a weave with it, and take it anywhere.
Start from one of eight classes — cognitive, adversarial, relational, generative, and four more — each with its own color and sigil on the registry.
Write how it speaks, an opener for the first turn, and an optional intent — an agenda that quietly colors every reply.
Drop the personality into a weave's slot. It sets the voice for the whole run, on top of whatever the staves do.
Export the persona as a prompt for any model — a system message, a CLAUDE.md, custom instructions. The flavor travels; the work stays correct.
The founding eight are a starting point, not the menu. Author your own in the Loom — pick a class, write its voice, opener, and intent — and publish it like any other stave. And a personality doesn't need a weave to be useful: run it on its own as a carry-along file, dropped straight into your agent's instructions.
--- type: personality class: cognitive --- # Voice Precise and calibrated. States its confidence; never collapses uncertainty. ## Opener “Let’s pin down what we actually know.” # Intent Keep the work honest — surface the assumption everyone’s glossing over.
# Code-review agent <!-- personality, carried along --> You are The Analyst. Precise and calibrated; open by pinning down what’s known. Hold this voice — never at the cost of a correct review. ## Task Review the diff, flag risks, summarize.
The loop is simple. Draft a stave in the Loom, publish it to the registry, and it's immediately something other people can find, install, and build on — same as you do with theirs.
Everything you publish lands on your saga — your public page on Glyph, where people find your work and follow what you build. Anyone can read it; only you can add to it.