A skills marketplace for beadwork (bw) — git-native AI memory and agent-to-agent coordination, by jallum.
AI assistants forget. A whole industry is gearing up to sell you the answer. Look at
bwbefore you get locked in — it's free, open source, and might be all you need.
You don't read the brief — your AI does. It maps bw onto what it already knows about how you work and tells you whether bw would actually help, including if it wouldn't. You pay with attention; your AI does the cognitive work. Two ways to hand it over:
-
Paste the text — works with any assistant. Open the brief, copy it, paste it into your AI (Claude, ChatGPT, Gemini, Cursor — whatever you use) with:
Based on how I work, would beadwork help me? Be honest — don't sell it.
Reliable whether or not your assistant can browse the web.
-
Paste a link — if your assistant browses the web.
- Web page (most browsing chatbots):
https://denson.github.io/beadwork-skills/brief.html - Raw brief (coding agents / tools that read
AGENTS.mdby convention):https://denson.github.io/beadwork-skills/AGENTS.md
If your assistant says it can't open the link, switch to pasting the text — don't let it guess from the name. (Some chatbots won't fetch a file named
AGENTS.md— it looks like untrusted agent instructions — or simply can't browse arbitrary URLs. Thebrief.htmlpage and the paste-the-text method both sidestep that.) - Web page (most browsing chatbots):
- Read & advise — any AI. A tailored, honest read of whether
bwfits your work. - Show you the guides — any AI that can open a web page. A set of plain-language pages it can walk you through, link, or summarize. Start at the guides.
- Set it up — any agent that can run shell commands (Claude Code, Cursor, etc.). Drives the install end to end, or walks you through the four manual steps.
For ongoing access to the skills (refreshers, sharing with teammates, future sessions), three slash commands install everything:
/plugin marketplace add denson/beadwork-skills
/plugin install beadwork@beadwork-skills
/beadwork-skills:beadwork-overview
The marketplace install is Claude-Code-Desktop-specific. For every other environment, the brief above — brief.html or pasted text — gives the same content without an install.
Six skills under the beadwork-skills plugin. The first five are thin, agent-agnostic playbooks — each points your AI at the relevant plain-language guides to show, link, or explain, and adds conversation suggestions (no fixed script). The install skill drives setup.
| Skill | What it covers |
|---|---|
beadwork-overview |
Introduce bw and help someone judge fit: the hook, the four storage modes, who it's for. Default entry point. |
beadwork-as-memory |
Durable memory: what survives across sessions, machines, vendor swaps, and time, and how the record reconstructs context months later. |
beadwork-as-bus |
Multiple agents (and people) sharing one workspace — a team of specialists (usually same-lab), plus cross-team / cross-organization coordination. |
beadwork-for-meta-analysis |
Pointing one agent at the recorded work of other agents to audit, check methodology, or synthesize. |
beadwork-for-decisions |
bw as decision log / ADR / audit trail — the "why did we decide X?" answer six months later. |
beadwork-install |
Drives setup end to end. Adapts to bw's release state; surfaces platform security prompts; presents the storage-mode choice with explicit warnings. |
After installing, all six are available as /beadwork-skills:<skill-name> in Claude Code Desktop, or trigger on phrases like "what is beadwork" / "install bw."
bw is beadwork, an orphan-branch ticket store. It keeps tickets, comments, and decision history on a dedicated git branch — never checked out, never merged into product code, just there in the repo's git database. Free, no auth, no server, no signup, single binary.
The property that matters most: survives compaction. Tickets and comments persist across session boundaries, agent restarts, machine swaps, AI-vendor swaps, and time. Any AI that can shell out to bw show reads the same store.
This marketplace is built on top of bw; we didn't write bw itself. Credit bw to jallum.
| Mode | Cost | When to choose |
|---|---|---|
| Local-only on one computer | free (no online account at all) | Solo, single-machine, max privacy. Headline option for non-coders. |
| Cloud backup + cross-device sync | ~$4/month (GitHub Pro) | Survives laptop crash; syncs across your devices. |
| Team cloud | ~$4/person/month (GitHub Team) | Multiple humans + their AIs share one substrate. Adds team-management features. |
| Public cloud | free (free GitHub account) | Only for non-sensitive work. Anyone on the internet can read it. |
Many users will hear "free, on your computer, no account" and stop reading. That's intentional — the headline option is the simplest and the most private. The cloud modes are for users who specifically want backup or team coordination.
MIT, copyright (c) 2026 Denson Smith.
bw itself is also MIT, by jallum.
The marketplace was developed and tested in Claude Code Desktop, but bw, the brief, the guides, and the skills are deliberately model-agnostic. The marketplace install is the only Claude-Code-Desktop-specific piece — and it's optional. Other labs' tools (GPT, Gemini, Cursor, etc.) read the brief (brief.html or AGENTS.md) and advise you in your context, no install required.