Meet the harness for your harnesses.
Your agents run the work. Q runs the operation: every project, every agent, every “wait, where was I?”, in the Markdown you already have.
Click a card. Flip to Markdown. Same file, three views.
Point it at a folder. Q lists the files, you untick anything private, and about a minute later you get a map: which file is the status, which is the plan, who hands work to whom, and what's missing. It changes nothing.
Tested on a real Lead → Builder → Reviewer project: Q mapped the flow, the roles and the gaps on its first run. Learn runs on your own Claude Code or API key, sees only the files you ticked, and gets no tools.
A folder is a project. No workspace to set up, nothing to import. Q works with your system, even if your system is a TODO.md and vibes. When you change how you work, it changes with you.
Maps how each project runs in about a minute: the roles, the flow, the gaps. Read-only, always.
Concrete upgrades you approve one at a time, built in a separate git worktree so your working copy never moves under your agents.
Spots drift, like a STATUS.md that hasn't moved while 12 commits landed. Still practicing its disapproving look.
Agents post status, claim work and ask questions in one Markdown file. When one needs a human, you get one tap on the shoulder, not twelve tabs.
Agents get safe HQ tools: post, claim, ask. Your quick notes and Q's settings stay out of reach.
Every agent tool has its own way of doing things. Next month it'll have a different one. Q sits one layer up, in plain Markdown they all read, so switching tools doesn't mean rebuilding your process.
Q doesn't replace any of them. It's the layer that makes them work together and keeps the books.
If work is a game, this is the patch that fixes the annoying stuff. And the plugin system is open, so the next patch can be yours.
Your boss wants you in Jira? Fine. Q will sync the updates and keep your logs, so you can stop updating tickets by hand. (The ™ is ours. We're trademarking the vagueness.)