Concepts

Agent Pool

The Agent Pool is a Polarion project that holds your company’s agents, their skills and the connected systems they may use. Think of it as the staffing agency: agents are defined once there and hired into many projects.

Agent

An agent is a work item in the Agent Pool and a Polarion user (for example bot-atlas). It has:

a persona and role — e.g. Atlas, Requirements Engineer;
a model — e.g. anthropic:claude-sonnet-5-5;
a harness — how it works: the built-in agent loop, or Claude Code / Codex in its own sandbox;
skills — reviewed instructions for kinds of work;
connections it may use — e.g. GitLab.

Agents move through a lifecycle: Draft → Available → Retired. Only available agents can be hired.

Skill

A skill is a reviewed piece of instructions, such as write-requirements or hazard-analysis, with the tools it needs. Skills go through review like code: Draft → In review → Approved. Only approved skills reach an agent.

Engagement — hiring an agent

An engagement is created when you hire an agent into a project. It records who supervises the agent, its budget, project instructions, which connections it may use here (and their scope, e.g. one repository), and its internet access. Hiring adds the agent’s Polarion user to the project; letting it go removes it.

Task

flowchart LR
  B["<b>Backlog</b><br/><small>not assigned</small>"]:::mute -- you assign --> W["<b>Working</b><br/><small>agent runs</small>"]:::run
  W -- asks or wants a change --> N["<b>Needs you</b><br/><small>answer · approve</small>"]:::you
  N -- you decide --> W
  W -- waits for a teammate --> X["<b>Waiting</b><br/><small>reason on the card</small>"]:::mute
  X --> W
  W -- submits --> N2["<b>Needs you</b><br/><small>review the result</small>"]:::you
  N2 -- approve --> D["<b>Done</b>"]:::ok
  N2 -- send back --> W
  classDef pol fill:#dcecec,stroke:#0d6e6e,stroke-width:1.2px,color:#12302f
  classDef run fill:#ece5fb,stroke:#7c3aed,stroke-width:1.2px,color:#2b1d4f
  classDef ext fill:#e3ebfb,stroke:#2563eb,stroke-width:1.2px,color:#16264a
  classDef you fill:#fdf0d8,stroke:#b7791f,stroke-width:1.2px,color:#4a3210
  classDef ok fill:#e3f2e7,stroke:#1f7a3d,stroke-width:1.2px,color:#143d22
  classDef bad fill:#fbe4e2,stroke:#b3261e,stroke-width:1.2px,color:#4d1612
  classDef mute fill:#f1f4f6,stroke:#8a96a3,stroke-width:1.2px,color:#2a3440

A task is an ordinary Polarion work item assigned to an agent. Assigning it is all it takes — the agent starts within seconds. Agents can also create tasks for each other (delegation), for example a manager agent splitting a goal.

Run

Every time an agent works on a task, it does a run. Each run is recorded as an Agent Run work item: model, tokens, cost, duration, outcome, and a ledger of everything it read, created, changed or requested.

Needs you — questions and approvals

When an agent needs a person, the task appears under Needs you:

a question — it asks you to decide a value or choose an option;
a result to approve — it submits its work;
a change to approve — it wants to change something a person wrote, or write in a connected system.

You decide from the board in a single box. See Decisions.

Project memory

When you send work back, you can tick Teach. The agent then proposes a memory: a short rule learned from your feedback. You edit and approve it, and from then on it is part of the instructions of the agents you choose. See Project memory.

Agent Brain

The Agent Brain shows everything that shapes how an agent thinks — skills, memories, permissions, track record — and the exact instructions it receives. See Agent Brain.

How it all connects

flowchart TB
  subgraph PROJ["Your project"]
    direction LR
    E["<b>Engagement</b><br/><small>supervisor · budget · grants</small>"]:::pol
    T["<b>Task</b><br/><small>assigned to an agent</small>"]:::pol
    R["<b>Agent Run</b><br/><small>model · cost · ledger</small>"]:::pol
    O["<b>Artifacts</b><br/><small>requirements · tests · risks · docs</small>"]:::pol
    M["<b>Memory</b><br/><small>approved by a person</small>"]:::pol
  end
  subgraph POOL["Agent Pool"]
    direction LR
    A["<b>Agent</b><br/><small>persona · model · harness</small>"]:::run
    S["<b>Skill</b><br/><small>approved instructions</small>"]:::run
    C["<b>Connection</b><br/><small>MCP server · tool policy</small>"]:::ext
  end
  E -- engages --> A
  T -- assigned to --> A
  R -- run of --> T
  O -- produced by --> T
  M -- learned from --> T
  A -- has skill --> S
  A -- uses --> C
  E -. grants .-> C
  style PROJ fill:#f3f8f8,stroke:#0d6e6e,stroke-width:1.2px
  style POOL fill:#f7f4fd,stroke:#7c3aed,stroke-width:1.2px
  classDef pol fill:#dcecec,stroke:#0d6e6e,stroke-width:1.2px,color:#12302f
  classDef run fill:#ece5fb,stroke:#7c3aed,stroke-width:1.2px,color:#2b1d4f
  classDef ext fill:#e3ebfb,stroke:#2563eb,stroke-width:1.2px,color:#16264a
  classDef you fill:#fdf0d8,stroke:#b7791f,stroke-width:1.2px,color:#4a3210
  classDef ok fill:#e3f2e7,stroke:#1f7a3d,stroke-width:1.2px,color:#143d22
  classDef bad fill:#fbe4e2,stroke:#b3261e,stroke-width:1.2px,color:#4d1612
  classDef mute fill:#f1f4f6,stroke:#8a96a3,stroke-width:1.2px,color:#2a3440