Founder and CEO of Vloora, a startup building the plumbing of the agentic web; Assistant Professor of Practice in Urban Technology, University of Michigan; Nonresident Senior Fellow, Global Office of Public Delivery.
Coalition is a Persistent Browser Based Game (PBBG) built in HTML5, the goal of the game is to escape before your character is captured.
In developing a text-heavy game with AI agents, I needed a system that kept in-game help text, narrative text and game rules all separate.
Based on design systems & style guides, I wanted a way to integrate my own editorial rules into the game while only having it impact the sections I wanted.
Every line comes from one catalog, not the code.
$ pbbg-content check content/pbbg/coalition.bundle.json warning COAL-10 ch2.level.eri_airstrip.fence.text: Owner, 2026-10-30: say a flight's length as '8-hour flight', never 'hours in the air'. error SPEAKER ch2.sibling_visa: Dialogue requires a speaker. Set speakerRef to a declared entity. warning TEXT-02 ch2.level.eri_town.bus.name: Review action labels longer than six words. $ python3 tools/build.py content-check: 0 errors, 0 warnings pbbg-check: Content checks passed.
The checker caught the first three lines when I added chapter two. The last two are the fixed build.
A simulation can run without errors while rewarding the wrong team. A model can recommend that an agent's request be reviewed without granting permission to act. A service can acknowledge a request without completing the job. I write about the rules, records, and checks that make those differences visible.
npm i red-pen-editorAn editor for AI agents with permission to cut. It looks for what a draft can lose without changing what it says, and it refuses any cut that would drop the last mention of a number, date, or link. Agents can use it as a skill or through an MCP server. It runs locally, with no API key.
npx -y github:quarterback/pbbg-content-schema install-skillA content format and checker that keeps a browser game’s rules, story, and interface text apart, so AI agents can edit one without breaking the others.
npm i closedtabShared memory for people and AI agents. It records what an agent decided and where it went off plan, which test results and agent logs often miss. The next session starts with that record.
Closedtab: A npm shared memory for you and your AI agents
Agent Experience Needs Failure Affordances
npm i prosekeepAn editorial system for teams writing with AI agents. It carries writing standards, author decisions, source requirements, and publishing workflows across repositories and agent tools. I treat editorial authority as a content design problem: an author’s decisions should still hold when the writing reaches editing and review in another tool.
npm i plain-language-defaultA reusable writing standard for AI agents based on the U.S. Federal Plain Language Guidelines. It gives every agent the same rules for writing plainly.