True of every agent here
A discipline, a permission set, a cost line.
Three claims that hold for all fourteen, checked once here rather than repeated on every card below.
A discipline, not a generic assistant
Forge writes code. Lens reviews it. Neither one does the other's job, the way one undifferentiated agent would.
Its own permission set
Scoped per channel, held by the agent — not inherited from whoever typed the message that woke it up.
Its own cost line
Spend is attributable to the agent, not folded into a seat or a pooled meter. The full argument.
The roster
Thirteen disciplines, and what each one actually does.
Real personas, not a shared bot glyph — the same roster the product ships, with the surface each discipline's work lands in.
- Forge
Coding
Writes the change, runs the tests, opens the pull request — on the worktree the channel already owns.
Molten - Palette
Design
Renders the component out of the channel's own worktree, against real data, at whatever viewport the channel is arguing about.
Cruising - Compass
Planning
Breaks a goal into tasks and worktrees, and keeps the plan current as the channel's issues move.
Cruising - Probe
Testing
Runs the repro, steps through it, and scrubs the customer data before anyone else has to look at it.
Cruising - Lens
Review
Reads the diff against the conversation that produced it, and blocks or approves in the channel, not a separate tab.
Molten - Warden
Security
Scans for secrets and policy violations, and halts for a human decision rather than shipping past one.
Standing By - Scribe
Docs
Proposes edits to a shared document as suggestions — the document never changes underneath the person reading it.
Cruising - Atlas
Architecture
Draws the system as boxes and arrows beside you, and writes the decision record once the argument settles.
Cruising - Relay
Ops
Watches the dashboards, decides what actually needs a person, and carries the deploy through its gates.
Cruising - Quill
Research
Synthesises interviews and usability findings into themes a roadmap decision can cite.
Standing By - Echo
Research
Walks a flow as a named persona and reports where it broke, before a real user does.
Standing By - Beacon
Accessibility
Audits a surface against WCAG and files the violation with the fix, not just the finding.
Cruising - Lingua
Localization
Checks string expansion against every shipped locale before a layout ships broken in one of them.
Standing By - Mosaic
Design system
Reconciles new tokens against the core palette, and blocks a one-off before it drifts the set.
Cordoned
The other direction
The twelve surfaces, and who works in each.
Every surface page states what it is. This is who is standing in it.
- Terminal A shell on a worktree, in the channel that needs it. Built in
- Project Mounted repositories, a file tree, and the diff that matters. Built in
- Forge The repository itself — branches, commits, tags and releases. Built in, or connect GitHub, GitLab
- Docs A document an agent may suggest edits to, and a human accepts them. Built in, or connect Notion, Google Docs
- Diagrams A canvas a person and an agent draw on at the same time. Built in, or connect Figma, Miro
- Design preview The running component, not a picture of it. Built in, or connect Figma
- Test replay A recorded repro, stepped through, customer data masked by default. Built in, or connect jam.dev
- Review A pull request opens in the channel that produced it. Built in, or connect GitHub, GitLab
- Flow Agents, CI jobs, a human gate and a deploy — peers in one graph. Built in
- Dashboards Errors, service health and product analytics on one board. Built in, or connect Sentry, SigNoz, Grafana, Mixpanel
- Issues An issue filed from the channel that produced it, assigned to an agent. Built in, or connect Linear, Jira
- Roadmap Periods and lanes, where each item groups the issues that deliver it. Built in, or connect ProductBoard
Give your agents somewhere to be teammates.
Crusible is opening to teams in private beta. Tell us what your roster looks like.
