Roles and guidance
Tell every assistant and agent how your project works — once, in pages they read before they act.
An assistant that joins your project should not have to guess how you name tasks, what a bug report looks like, or what “done” means here. Rindo keeps those instructions as pages in the project — one for the project, one per work role, one per skill — and makes them available to every assistant and agent: to read over MCP, to attach in Claude, or as references on an agent’s job.
Guidance
Every project has a Guidance page: the conventions of the project — task naming, the bug report format, the workflow, what a reviewer looks for. Find it under Documents → Guidance.
- Any assistant connected to the project, and any project agent, can read it before working; over MCP it is one call away, and in Claude it can be attached to a message from + → Connectors → Add from your Rindo connector.
- Admins edit it. An agent can only propose a change to it, as a draft an Admin approves.
Work roles
A work role names a job your project needs done — BA, QA, reviewer, translator — and carries its own instructions. It is not a permission: a Viewer can hold a role, and holding one grants nothing beyond what the holder’s permission already allows. The product keeps the three words apart — Permission is the level (Owner · Admin · Member · Viewer), Scope is what a token or agent may do, Role is the job.
Define roles under Settings → Roles: a short immutable key, a name and a required description. Rindo creates the role’s instructions document with it — an ordinary page in the project that Admins edit (agents propose drafts). Up to fifty roles per project.
Assign holders from the Members table (the row’s ⋯ Roles…) or from an agent’s edit drawer: a holder is a person or a project agent. Removing a member or an agent clears their roles, and every role change is on the audit trail.
Where the instructions reach
| Holder | How the role’s instructions arrive |
|---|---|
| A person using Claude | the connector lists one prompt per role you hold, named <project key, lowercased>-<role key>; picking it attaches the instructions to your next message (a new conversation needs the pick again). Download plugin in the role row’s ⋯ menu — shown only to a holder — saves a zip to upload through Claude’s Upload plugin |
| A person using Claude Code | the instance’s plugin marketplace lists a rindo-role-… plugin for each role you hold |
| A project agent | the agent’s job carries references to its own instructions document and to every role it holds, in order; the agent reads each one over MCP, and the runner itself never opens them |
A project agent can also carry an instructions document of its own, independent of roles, set in the agent’s drawer.
Skills
A skill is a packaged, reusable instruction for coding agents — a SKILL.md with optional text files — kept as a page in the project and listed under Skills. Skills are published deliberately: approving an edit is not publishing it, and only the published revision is served. Before each job, a runner syncs the project’s published skills into ~/.claude/skills, where Claude Code finds them natively. Admins edit and publish; agents propose drafts.
Libraries
A Library is a project whose Guidance, published skills and pages are packaged as a versioned pack for other projects to use. A project Admin binds a Library under Settings → Packs, pinned to a version; its pages and skills then appear in the project as inherited, read-only content, and the Admin can hide single items or advance the pin when a new version is published.