Skip to content
Developer Preview — known limitations are marked in red and will be fixed for production.

Skills

A skill is a set of instructions an agent follows, plus the list of Polarion tools those instructions need. Think of it as a reviewed procedure: “how we write requirements here”, “how we plan work”, “how we analyse hazards”. Agents get their abilities only from skills, so skills are where you decide what an agent may do in Polarion.

Skills are work items of type Skill in the Agent Pool. An agent has a skill through the has skill link.

A skill work item: approved status, its tools and the instructions agents receive

Part Where Meaning
Name Title Short name, e.g. write-requirements.
Instructions Description What to do and how, in plain language. Headings and lists are fine.
Tools Tools field The Polarion tools this skill allows.

Instructions become part of the agent’s system instructions. You can see the exact result for any hired agent on its Agent brain panel (see Agent brain).

stateDiagram-v2
  direction LR
  [*] --> Draft
  Draft --> InReview: submit
  InReview --> Approved: approve
  InReview --> Draft: send back
  Approved --> Draft: rework
  Approved --> Retired: retire
  Retired --> Draft: rework
  InReview: In review
  classDef ok fill:#e3f2e7,stroke:#1f7a3d,color:#143d22
  classDef you fill:#fdf0d8,stroke:#b7791f,color:#4a3210
  classDef mute fill:#f1f4f6,stroke:#8a96a3,color:#2a3440
  class Approved ok
  class InReview you
  class Draft,Retired mute

A skill is code in plain language: it changes how an agent behaves on every task. That is why skills have a review workflow, and why only approved skills reach an agent. A draft or a skill in review is ignored, even when it is linked to an agent.

Status Meaning Actions
Draft Being written or changed. Not used. Submit for review → In review
In review Waiting for a reviewer. Not used. Approve → Approved, Send back → Draft
Approved Used by every agent that has it. Rework → Draft, Retire → Retired
Retired Out of service. Not used. Rework → Draft

To change an approved skill, run Rework, edit it, then submit and approve it again. While it is in draft or review, agents work without it, so plan bigger changes for a quiet moment or write the new version as a separate skill.

The tools field lists Polarion tools by id. Some tools come with every agent, some are unlocked by others.

Always available to every agent: read_work_item, query_work_items, list_documents, read_document, describe_type, add_hyperlink (links to objects in connected systems) and propose_memory (see Memory).

Selectable in a skill:

Tool Lets the agent
create_work_item Create work items. Also unlocks editing: updating fields, linking, test steps, editing and creating documents, recording test runs, requesting reviews, workflow actions on its own items, attaching files, commenting on documents.
link_work_items Link work items.
update_field Change a field.
comment Comment on work items.
delegate Hand sub-tasks to teammates. Also unlocks reviewing their results and answering their questions.
ask_human Ask the supervisor a question.
submit_for_approval Submit a result for the supervisor’s approval.
add_image, set_images Add images (for example design screens) to items.

The Tools field offers further ids that are included with create_work_item anyway, such as create_document, update_fields or record_test_run. See the tool reference for every tool.

Whatever a skill allows, some limits always apply:

  • Changes to items written by people are held for the supervisor’s approval. See Governance.
  • Agents may run workflow actions only on items they created, and never into states such as approved, done, verified, closed, rejected or obsolete, and never actions that need a signature.
  • Tools of connected systems are not granted by skills. They come from connections.
  • Keep each skill about one kind of work. Give agents several skills rather than one large one.
  • Say what “done” means: which items to create, which links to set, when to ask and when to submit.
  • Name the work item types and link roles of your projects, so the agent doesn’t guess.
  • Write rules, not examples tied to one project. Project specifics belong in the engagement’s project instructions, which a project lead sets when hiring.
  • Give a skill only the tools it needs.

Skills in the Agency page: status, tools and the agents that use them

  • Agency → Skills lists every skill with its status, number of tools, the agents that have it and when it last changed.
  • Each agent run record lists the skills the agent ran with, so you can tell which instructions were in force for any result.