Skip to main content
A skill is a named set of instructions the AI Agent loads when a task calls for it. It tells the agent which tools to call, in what order, and where to stop and ask. Skills live in your workspace, so you write a procedure once and every chat can use it. This page covers how skills are structured and how to write your own. For using the agent day to day, see AI Agent.

​
The skills library

Open AI Agent from the left rail and select Skills.
AI Agent Skills page with callouts for the filter, New skill, Use in chat, and duplicate buttons, and for the Usage and History pages in the sidebar.AI Agent Skills page with callouts for the filter, New skill, Use in chat, and duplicate buttons, and for the Usage and History pages in the sidebar.

The skills library

Each skill carries a badge that shows who maintains it and who can see it: Two curated skills ship today:
  • rtlca-map-material — manual, user-approved mapping of inventory materials to EPDs. See Map materials to EPDs.
  • rtlca-compare-relevant-epds — finds alternative EPDs for a material and ranks them by project GWP impact. See Compare EPDs.

​
Anatomy of a skill

Open any skill to read it in full. Curated skills are read-only, but nothing is hidden — the page shows the complete instructions the agent receives.
Detail page for the rtlca-map-material skill showing its description, when-to-use text, and the rendered instructions.

A curated skill, in full

A skill has four fields: The Skill body is where the work happens. Write it as instructions to the agent: name the tools it should call, the checks to run before writing data, and the point where it must stop for the user’s approval.

​
Write your own skill

The fastest start is to duplicate a curated skill and adapt it.
1

Duplicate a skill

Click the duplicate icon on a skill card, or Duplicate on its detail page. You get a private copy named <skill> (copy) with every field pre-filled.
Duplicate Skill form pre-filled with the name, description, usage instructions, and body of rtlca-map-material.
To start from a blank form instead, click New skill.
2

Edit the fields

Rename the skill and adjust the description and body. Keep the description specific — it is what the agent reads to decide whether the skill fits.
3

Choose who can see it

Tick Share this skill with everyone in the organization to publish it to your team. Leave it unticked to keep it private.
Bottom of the Duplicate Skill form with the share-with-organization checkbox and the Create button.
4

Click Create

The skill appears in the library with a Private or Shared badge. Skills you own also get a delete button.
Skills library listing a private copy, a shared copy, and the two curated skills.

​
Writing guidelines

  • Reference tools by their exact name. The curated skills call rtlca_map_materials and rtlca_compare_epds by name, so the agent never has to guess.
  • Gate every write. Tell the agent to present its plan and wait for approval before any tool that changes project data. The agent still asks for Approval needed on writes, but the skill should make the decision clear before it gets there.
  • State the preconditions. The mapping skill checks for a non-zero quantity, a set lifetime, and a compatible declared unit before it maps anything. Spell out the equivalent checks for your task.
  • Say what not to do. rtlca-map-material forbids the automapping tools outright. Closing off the wrong path is as useful as describing the right one.
  • Define the output. Ask for a compact table, a one-line confirmation, or a short summary, so results are consistent across runs.

​
Run a skill

Attach a skill from Skills in the composer, or click Use in chat in the library. The skill shows as a chip above your message.
AI Agent panel on a project dashboard with a Using: rtlca-compare-relevant-epds chip above a prompt asking for alternatives to a concrete.

A skill attached to the composer

The agent follows the skill’s procedure and shows each tool call and its reasoning as it goes.
Agent trace showing several completed rtlca_compare_epds calls with an expanded thinking step between them.

Tool calls made by the skill

Agent answer listing relevant EPD alternatives ranked by project GWP impact in a table.

The result, formatted as the skill specifies

To see how much your skills consume, use Usage and History in the same sidebar. Both are described in AI Agent.

​
Use skills outside Real-Time LCA

A skill body is plain Markdown, so it also works in Claude, ChatGPT, or Le Chat connected through the Real-Time LCA MCP. Copy the body from the skill’s detail page, or take a ready-made prompt from the MCP use cases.