Role Management

Role Management

Intro

A Role is a named set of responsibilities that an Actor can fill. Roles describe what — not who. "Alice is a developer" is a Binding between Actor ACTOR-alice and Role ROLE-developer; this skill manages the Role side of that relationship.

Important: Roles in processkit have no enforcement semantics. They are descriptive, not restrictive. RBAC enforcement is out of scope (see DISC-002 / DEC-017).

MCP server. This skill ships a self-contained MCP server at mcp/server.py (PEP 723 script — requires uv and Python ≥ 3.10 on PATH). Agent harnesses reach its tools by reading a single MCP config file at startup, so the contents of mcp/mcp-config.json must be merged into the harness's MCP config and placed at the harness-specific path before this skill is usable. If processkit was installed by an installer, that wiring is the installer's responsibility; if processkit was installed manually, the project owner must do it by hand.

Overview

When to create a Role

Create a Role when:

• A responsibility recurs across multiple work items or processes.
• You need to refer to "the person who X" without naming a specific actor.
• A process template references the role ("the reviewer must …").

Shape

---
apiVersion: processkit.projectious.work/v2
kind: Role
metadata:
  id: ROLE-reviewer
  created: 2026-04-06T00:00:00Z
spec:
  name: reviewer
  description: "Reviews code and documentation changes before merge."
  responsibilities:
    - "Read PRs within 48 hours of request"
    - "Approve or request changes with actionable feedback"
    - "Block merges on security or correctness concerns"
  skills_required: [code-review, debugging]
  default_scope: project      # project | sprint | permanent
---

Optional body: elaboration, examples, anti-patterns.

Workflow

1. Pick an ID: ROLE-<name> where name is kebab-case and unique.
2. Write a one-sentence description — the role's purpose.
3. List concrete responsibilities as imperative bullet points.
4. Optionally list skills_required — links to skill IDs that describe the capabilities this role depends on.
5. Save to context/roles/.
6. Log role.created.

Assigning actors to roles

Do not list actor IDs in the Role file. Assignments go in Binding entities so they can be scoped, time-bounded, and changed without editing the Role.

kind: Binding
spec:
  type: role-assignment
  subject: ACTOR-alice
  target: ROLE-reviewer
  scope: SCOPE-project-x    # optional

See skills/binding-management.

Gotchas

Agent-specific failure modes — provider-neutral pause-and-self-check items:

• Confusing Role with Actor. A Role is a named set of responsibilities ("tech-lead", "release-manager"). An Actor is a person or agent who can FILL a role. "Alice is the tech lead" is a Binding from ACTOR-alice to ROLE-tech-lead, not an update to either entity.
• Creating a Role for one specific person. "Alice's responsibilities" is not a Role; it's just Alice. Roles capture responsibilities that can be transferred between people. If only one person could ever fill it, it's a job title, not a Role.
• Treating Role membership as RBAC enforcement. Roles are descriptive, not restrictive. processkit does not check whether an actor is "allowed" to do something based on their Role. Authorization is out of scope — Roles document who SHOULD do what, not who is permitted to.
• Renaming a Role instead of creating a new one. Renaming silently rewrites history — every old Binding now points at a Role with a different meaning. If the responsibilities have genuinely changed, create a new Role and end the old Bindings; don't overwrite.
• Putting tasks in responsibilities instead of WorkItems. The responsibilities field describes ongoing duties ("review PRs", "approve releases"), not specific to-dos. If you find yourself writing "implement feature X" in responsibilities, that's a WorkItem.
• Vague responsibility lists. "Ensures quality" is not a responsibility; "Reviews every PR before merge to main" is. Each bullet should be concrete enough that you could ask a holder "did you do this last week" and get a yes/no answer.
• Forgetting to log role.created / role.updated. The audit trail tracks role evolution. Skipping the log makes "when did we add the security-reviewer role" unanswerable.

Full reference

Fields

Roles vs Categories vs Gates

Beginners sometimes conflate three related concepts:

• Role — who does a thing (reviewer, developer, release-manager).
• Category — how a thing is classified (bug, feature, tech-debt).
• Gate — a checkpoint that validates a thing (code-review-passed).

A Role can be assigned via Binding; a Category is a label; a Gate is something a process has to pass. Don't model one as another.

No enforcement

processkit does not check that an actor filling a role has the right skills_required, or that actions restricted to a role are only performed by bound actors. Roles are descriptive. If you need RBAC, that's a governance platform concern (DEC-017).

Well-known roles

processkit does not ship a canonical list of roles — they vary too much per project. But these are common enough to be mentioned here for naming consistency: developer, reviewer, tech-lead, release-manager, incident-commander, product-owner, designer, security-reviewer. Projects are free to use these names or invent their own.