Skills save the way you complete a recurring piece of analysis, so the next daily pick-up check or weekly performance brief starts with a proven process instead of a blank chat. A Skill can ask for the details it needs, query current Insights data, work through several steps, and return the result in a consistent format whenever you run it.
What a Skill can do
When to use a Skill
Creating your first Skill
Writing reliable instructions
Running and refining a Skill
Testing a Skill
Current limits and visibility
Examples and templates
What a Skill can do
A Skill is a named set of instructions for the Agent. It is more capable than a saved one-line prompt: the instructions can describe a complete workflow and the Agent works through that workflow in chat.
A Skill can:
Ask for missing information, such as a property, stay period, comparison period, or threshold.
Run one or several queries, including queries from different Insights Topics when the analysis requires them.
Apply definitions, checks, thresholds, and decision logic consistently.
Combine current Insights data with a file you attach for that run, such as a corporate-account list or reporting template.
Return a written briefing, table, visualisation, or dashboard, depending on the workflow you describe.
Handle missing or ambiguous data in the way you specify instead of guessing.
The Skill provides the process; the Agent still works with the current data and access available in your Insights session. Each run is a fresh analysis, so the result can reflect the latest available data.
When to use a Skill
Use a Skill when the job has a repeatable shape: the same questions, calculations, checks, or output format each time, even if the property or date range changes.
Use a Skill for an on-demand workflow you expect to repeat, such as a morning pick-up briefing or monthly corporate-account review.
Use chat when the question changes direction as you investigate and you do not yet know what the final process should be.
Use a dashboard when you need an always-current, multi-page view that people can revisit without running a new conversation.
Use a scheduled dashboard when the output must be refreshed and delivered automatically by email or another destination.
A Skill runs only when you start it. Saving a Skill does not schedule it, monitor your data in the background, or email the result automatically.
Creating your first Skill
There are three ways to create a Skill.
Create with the Agent
This is the easiest place to start. You can turn a useful conversation into a Skill, or describe the recurring job directly.
Open the Agent.
Describe the workflow and include the property or date inputs, metrics, comparisons, and output you need.
Ask the Agent to turn it into a Skill.
Review the proposed name, identifier, description, and instructions before relying on it.
Giving the Agent the full brief in one message reduces unnecessary back-and-forth. For example:
Create a Skill called Weekly segment pick-up. When it runs, ask once for the property if I have not supplied one. Use the Pick-Up Topic to review the last seven days of booking activity for stay dates in the next 90 days. Show Units Pick-up, ADR Pick-up, and Accommodation Revenue Pick-up by segment in a table, followed by three concise findings. If no results are available, say so rather than estimating. Before creating the Skill, show me the proposed fields and instructions so I can review them.
If you have already completed the analysis in chat, you can simply say: Turn this conversation into a Skill. Check that the saved instructions preserve the decisions and corrections you made during the conversation.
Write your own
Open Skills from the left-hand menu.
Select New skill and then Write skill instructions.
Complete the four Skill fields.
Select Create skill.
Upload Skill instructions
Select New skill and then Upload a skill to import an existing Markdown Skill file. Insights can read the name, identifier, and description from optional information at the top of the file, then use the remaining Markdown as the instructions.
This is different from attaching reference data when you run a Skill. Upload a skill imports the Skill definition; Attach file gives the Agent additional context for the current conversation.
A Markdown file can use this structure:
--- identifier: /weekly-pickup name: Weekly pick-up description: Reviews recent pick-up for one property and highlights the main contributors. --- [Write the Skill instructions here]
Understanding the four Skill fields
Name — the display name on the Skills page. Keep it to two to four words that make the job obvious. Maximum: 48 characters.
Identifier — the command used to run the Skill, such as
/weekly-pickup. Use lowercase letters, numbers, and dashes, and begin with a letter or number. Maximum: 48 characters. Identifiers must be unique; if yours is taken, add a meaningful distinguishing word such as a team or property code.Description — a short summary shown when you choose the Skill. Say what it produces and any important input it requires. Maximum: 1,024 characters.
Instructions — the process the Agent follows, written in Markdown. Maximum: 20,000 characters.
Writing reliable instructions
Start with the smallest set of instructions that describes the job clearly. Run the Skill, observe where the result is inconsistent, and add precision only where it improves the outcome.
For most analytical Skills, this structure is a useful starting point:
PURPOSE Explain the one recurring job this Skill should complete and why the output is useful. INPUTS Ask once for any missing property, stay period, comparison, threshold, or reference file. Use information already included when the Skill was started. DATA SCOPE Name the relevant Insights Topic, time field, date range, metrics, and groupings when they matter. STEPS 1. Run the first part of the analysis. 2. Run any required comparison or check. 3. Bring the findings together. OUTPUT Specify the table, chart, headings, narrative length, and sorting you want. GUARDRAILS Explain what to do with missing, ambiguous, or zero-result data. Tell the Agent not to invent unavailable figures. FINAL CHECK Ask for the bottom line, the evidence behind it, and any limitation that affects confidence.
A few principles make this structure work well:
Give the Skill one clear job. Separate unrelated analyses into focused Skills.
Define a metric when an ambiguous definition could change the result. You do not need to redefine every familiar term.
Explain why an important rule exists. This helps the Agent apply the intent when the exact situation changes.
Specify what a successful result contains, including the order and format.
State how to handle missing data and empty results so the Agent reports the limitation rather than manufacturing a finding.
Avoid repeating the same instruction in several places. Longer instructions are not automatically better.
Running and refining a Skill
You can run a saved Skill in three ways:
Type
/in chat and choose the Skill, or type its identifier directly.Open the Skill from the Skills page and select Run.
Select the + icon beside the chat input, choose Skills, and select the Skill.
You can add a condition when you start the Skill without changing its saved instructions. For example:
/weekly-pickup for Flagship Austin by Flyr, limited to the next 30 stay dates
The Agent applies your additional request together with the saved instructions.
If a Skill needs a reference file, write the instructions to ask for it and attach the current file when you run the Skill. The file provides context for that conversation; it is not saved as part of the Skill for future runs.
To edit a Skill in place, open it from the Skills page and edit the fields directly. Asking the Agent to revise an existing Skill in chat currently creates a new Skill instead of updating the original.
Testing a Skill
Test a new or significantly changed Skill before relying on it for a recurring report. Run at least these three cases:
Normal case — supply the usual property and period. Check that the correct Topics, metrics, comparisons, and output format are used.
Missing or ambiguous input — leave out the property or use an unclear name. Check that the Agent asks once for what it needs and does not guess.
No-data or edge case — choose a period with no activity or a property without the optional comparison data. Check that the Agent explains what is missing and does not invent a result.
If a test fails, correct the smallest instruction responsible and run all three cases again. This produces a more flexible Skill than adding broad restrictions after every unexpected result.
Current limits and visibility
A Skill is on-demand and runs only when you start it.
Your Skills appear only on your own Skills page; they are not currently shared with teammates.
An identifier must be unique, even though the Skill itself is private to you.
A Skill cannot use information that is not available in Insights unless you provide that information in the current conversation.
A Skill can improve consistency, but its response is still generated by the Agent. Review important figures and conclusions before using them in a decision or sharing them externally.
Examples and templates
See Skill Examples and Templates in Insights for five copyable examples covering a weekly pick-up check, a guided forecast, a GM brief, a file-assisted corporate-account review, and a data-quality check.
Have questions or run into something unexpected? Reach out anytime via the in-app chat — we're happy to help.