ICSE 2027 · Tool demonstration

Myzelium: steering AI coding agents and the people behind them with a typed spec graph

This page accompanies the tool paper. Sign in with one of the reviewer accounts below. No installation is needed. The workspace holds a synthetic example project, Pocket Wallet (a mobile wallet with 8 epics and 88 user stories), so you can follow the paper's workflow end to end.

Reviewer accounts

EmailPassword
[email protected]icse27-rmmm-nbl6
[email protected]icse27-4utv-hi9q
[email protected]icse27-izyh-vkni

Any account works; using different ones keeps your Clarity conversations apart.

A ten-minute tour

  1. 01
    Configure

    Settings → Spec model shows the tenant's types: an Epic and User Story ladder on top of the built-in System build contract, with their kinds, phases and the agents bound to each.

  2. 02
    Seed

    Open Pocket Wallet. Ask Clarity (bottom right) for a new story, e.g. “Add a story to freeze my card from the app”. It proposes a plan you review before anything is written. The project also holds one meeting-notes document, “Risk & compliance review”.

  3. 03
    Refine

    Open the story “Split a bill with friends”. It is deliberately weak: score it, answer one of its routed questions, and let Clarity fold the answer in and re-score it.

  4. 04
    Build

    Plan → Optimized route shows the computed build order, its batches and the saving against the roadmap order. “Copy my wallet address to the clipboard” is a small, ready story if you want to start a Dev Session.

  5. 05
    Keep it alive

    The meeting notes cap peer-to-peer transfers at $500 a day; the story “Send money to a contact” says $2,000. The project overview's Findings list that contradiction and the gaps the meeting raised.

  6. +
    Bring your own agent

    Profile → MCP creates a key and shows the exact command to connect Claude Code, Cursor or Claude Desktop to the same graph. Then ask your agent what to build next in Pocket Wallet.

About this workspace

  • The three accounts share one project, so you may see other reviewers' edits.
  • Agent runs (Dev Sessions, VirtualQA, VirtualDesigner) are capped: one at a time and a fixed number per month across all reviewers.
  • Meeting bots and email invitations are switched off in this workspace.
  • No personal data is needed or collected: the accounts are shared and the project is synthetic.