Introducing Quiddity: generate essential skills for your dev workflow

Over the past few weeks I've written a series on building agent skills for dev workflow — a /next-task skill that picks up work and opens PRs, an /approve skill extracted from shared patterns, and a /new-issue skill that files structured issues from a sentence. Together they form a build loop: file an issue, pick it up, ship it, repeat.

The catch is that every team's loop looks different. A /next-task skill that works great with Linear and GitHub is useless if your team runs Jira and GitLab. Branch conventions, CI setup, review process, merge strategy — all of it varies. These skills have to be bespoke. After building my own set by hand and watching the patterns repeat, I realized the setup itself could be automated.

So I built Quiddity.

The name comes from Medieval Latin quidditas, meaning "whatness" — the essential nature of a thing. That's what Quiddity does. It figures out the what of your project — your tools, your process, your conventions — then generates agent skills tailored to how you actually work.

One command to install, one command to set up

Getting started is two steps:

npx skills add jlong/quiddity

npx skills add jlong/quiddity

Then run /q-setup in your agent. Quiddity walks you through a conversation: it scans your project, interviews you about your development process, identifies the tools you use, helps install any missing CLIs or MCPs, and generates three custom skills — /new-issue, /next-task, and /approve — wired up to your specific stack.

What a setup session looks like

Here's an abbreviated session:

> /q-setup

Scanning project structure...
✓ Wrote .quiddity/project.md

Do you have a document that describes your development process?
> No, just interview me.

What branching model do you use?
> GitHub flow — short-lived feature branches off main.
...
✓ Wrote .quiddity/process.md

Which issue tracker do you use?
> Linear, team is ENG
Which source control platform?
> GitHub
...
✓ Wrote .quiddity/tools.json

Checking tool setup...
Tool             CLI            MCP
GitHub           gh (ok)        —
Linear           linear (ok)    Linear MCP (not found)

Linear MCP lets the agent create and update issues directly.
Want me to set it up? [Y/n]

✓ Generated skills/new-issue/SKILL.md
✓ Generated skills/next-task/SKILL.md
✓ Generated skills/approve/SKILL.md

> /q-setup

Scanning project structure...
✓ Wrote .quiddity/project.md

Do you have a document that describes your development process?
> No, just interview me.

What branching model do you use?
> GitHub flow — short-lived feature branches off main.
...
✓ Wrote .quiddity/process.md

Which issue tracker do you use?
> Linear, team is ENG
Which source control platform?
> GitHub
...
✓ Wrote .quiddity/tools.json

Checking tool setup...
Tool             CLI            MCP
GitHub           gh (ok)        —
Linear           linear (ok)    Linear MCP (not found)

Linear MCP lets the agent create and update issues directly.
Want me to set it up? [Y/n]

✓ Generated skills/new-issue/SKILL.md
✓ Generated skills/next-task/SKILL.md
✓ Generated skills/approve/SKILL.md

Quiddity stores its configuration in a .quiddity/ directory — project.md, process.md, and tools.json. Commit these alongside the generated skills and your whole team shares the same setup.

Why bespoke matters

Generic automation breaks the moment your process deviates from the template. A hardcoded Linear integration doesn't help a team on Jira. A skill that expects main as the default branch fails if yours is develop. The whole point of using an LLM for workflow is that it adapts to your context instead of forcing you into someone else's.

Quiddity leans into this. Instead of shipping a one-size-fits-all skill, it asks you how things work and writes skills that match. Your branch naming conventions. Your commit message format. Your issue tracker fields. Your CI checks. The generated skills are plain Markdown files you can read, edit, and version — they're not a black box.

A precursor to the loop

Once your skills are dialed in, they're also a natural precursor to a Ralph loop — an autonomous bash loop that runs your agent repeatedly until the work is done. File issues into the backlog, kick off the loop, and let it churn through tasks while you focus on design, architecture, or whatever needs a human eye. The skills Quiddity generates give the loop the structure it needs to operate safely.

Try it

Install the skills, run /q-setup, and you'll have a working dev workflow in a few minutes. The whole thing is open source on GitHub.

npx skills add jlong/quiddity

npx skills add jlong/quiddity

Start with one project and tweak as you go.

Quiddity is early and I'd love help shaping it. Try it on your project and let me know what works, what breaks, and what's missing. What tools and workflows would be most valuable to support next? Open an issue or reach out on X — I want to hear what your dev workflow looks like.