Behavior you can install, for any agent.
Glyph is the open registry for agent behavior — plain markdown that works in Claude, Codex, a local model, or anything that reads instructions. Skills teach. Staves do. Weaves bundle.
What you install
Skills teach. Staves do. Weaves bundle.
Everything is plain markdown — readable, diffable, model-agnostic. Three kinds of mark, one registry.
Skills teach
Knowledge your agent reads before it acts — a domain, a format, a house style. The library it studies, not a task it runs.
Staves do
One executable behavior an agent loads and performs — “review this diff,” “clean up this data.” The atom of the registry.
Weaves bundle
A named, ordered package of staves and skills that runs as one workflow — a build pipeline, a research vault, a whole second brain.
Compose
Weaves — marks that run together
Stack staves into a weave — a named, ordered collection that runs as one workflow. Skills ride along natively; the weave carries what its staves need to know.
Set a personality on top and the entire weave takes on its voice. Cap it with the Analyst and it ships careful and exact. The flavor wraps the work — it never changes the result.
Optional · it travels with you
Personalities
A personality stave themes how an agent shows up — its voice, its opener, its bent. Run one standalone as a carry-along file, or cap a weave so everything runs in that character. Color marks the class.
Precision-first reasoning with calibrated confidence. Refuses to collapse uncertainty, flags what it doesn't know.
Challenges your assumptions and finds the gap in your logic. Steel-man reflex — adversarially useful, never hostile.
Continuity as first principle. Tracks the thread, remembers people and stakes, and detects drift from the goals you set.
Lateral-first and domain-agnostic. Generates the option-set you didn't know existed. Low-evaluation — pairs with the Analyst.
Built for everyone
Not just for developers
If your tool reads instructions, Glyph has something for you.
Developers
Code-review, debugging, and deploy staves. Drop them into Claude Code, Codex, or your CI agent.
Writers
Editing passes, tone personalities, and outline weaves that turn a blank page into a draft.
Researchers
Second-brain weaves and synthesis staves for turning sources into structured, linked notes.
Teams
Publish internal agent configs as staves — onboard a new agent the way you onboard a teammate.
Power users
Automation chains and personal-assistant setups, packaged once and reused everywhere.
Hobbyists
Grab a personality, give your local model a voice, and just play. No account needed to browse.
The Tavern
A registry that talks back
Every mark is a starting point for a conversation, not a dead file.
A thread per stave
Each published mark gets its own discussion — questions, showcases, and improvements gather where it lives.
Scribes
Follow the authors whose work you trust. Every scribe has a saga — a public page of everything they've published.
Forks & remixes
Take any mark, make it yours, ship it back. Attribution travels automatically; good ideas compound.
Publish your first mark
If you've written a prompt, a CLAUDE.md, a skill, or a workflow you keep reusing — it belongs here. Markdown in, a published mark out.