How to Run a Skill
Last updated: March 9, 2026
To run a Skill, tell Ellie which Skill to use and provide the relevant documents. She will handle the rest and show the results when complete.
Skills can also be run directly from the Skills page.
How to Trigger a Run
Tell Ellie which Skill to run and with which documents — for example, "Run the Quarterly Rent Roll Analysis on the January rent roll I just uploaded." If documents are already attached to the conversation, Ellie will use them automatically.
Extra instructions can be provided for a specific run without changing the Skill itself. For example: "Run the Lease Abstract Summary but focus on the renewal terms this time." This tailors the output for that run only — the underlying Skill stays the same for next time.
The rule of thumb: if the change is needed for this run only, add it as an instruction. If it should apply to every future run, edit the Skill.
Document Guidance
It is recommended to use a focused set of documents per run for the most thorough and consistent results. The platform can examine each document in depth and cross-reference between them. With more documents, the Skill can still produce good work, but may not examine every page at the same depth. If more documents are provided than the Skill can handle in a single run, Ellie will ask which ones to focus on before proceeding.
For processes that regularly involve a large number of documents — say, abstracting 30 leases from a new acquisition — it is recommended to split into batches across multiple runs. Running the Skill several times on smaller groups of leases is more reliable than processing them all at once.
💡 Tip: If a process involves more documents than a single run can handle, split them into focused batches. Several smaller runs will produce more thorough results than one large one.
These limits are expanding as the platform adds new capabilities — particularly Skill composition, which will give each sub-step its own full context window.
What Happens During Execution
Skills run as background work. Depending on complexity and document count, execution takes anywhere from a few seconds (simple extraction from a single lease) to several minutes (multi-step analysis cross-referencing several sources). There is no need to stay on the page while it runs.
The platform has built-in error handling. If something unexpected occurs — missing data, an unexpected format — the system attempts to resolve it automatically before producing output.
After a Run Completes
Output appears in the conversation thread — either as text, a generated file (DocX, CSV, or Excel), or both. Generated files include an inline preview for reviewing without downloading.
Ellie will explain the output and ask if it matches what was expected. If anything needs adjusting, she can help refine the Skill for future runs.
Run Skills One at a Time
It is recommended to run Skills sequentially — review the results from one before starting the next. This ensures each Skill has the right inputs and output can be verified before moving on.
This is especially important when Skills build on each other. For example, running a rent roll extraction Skill first, reviewing the output, and then feeding the results into a variance analysis Skill makes the second step more reliable.
💡 Tip: A specific run can be tailored by adding extra instructions — e.g., "focus on the renewal terms this time." This adjusts that run only without changing the Skill itself. To make the change persist, edit the Skill instead.