Atlassian Rovo Chat now enables you to turn repetitive work into Custom Skills.
For example, if every week you prompt AI:
“Organize this week’s Jira: find completed items, blockers, next week’s priorities, and create a summary easy for managers to understand.”
If that prompt already works well,
it’s tempting to think:
“I’ll just save the entire prompt as a Skill.”
But don’t do that yet.
Because a “prompt that works today”
doesn’t necessarily mean “a process safely repeatable next month.”
Today you’ll learn one method:
Before building a Custom Skill,
break the content down into:
Fixed Rules / Variable Inputs / Mandatory Human Confirmation.
① Fixed Rules: Only include what really stays the same every time
Start by asking:
“If the data changes next week, does this rule still apply?”
If the answer is “yes,”
that rule fits in the Skill.
For example, every weekly Sprint Summary includes fixed requirements:
- List completed items of the week
- Blockers in a separate section
- Next week’s priorities capped at 5 items
- No adding dates that don’t exist
- Use bullet points
- Highlight any missing data items at the end
These rules don’t suddenly break just because:
Sprint 23
changes to
Sprint 24
That’s why they form the:
framework of the Skill.
② Variable Inputs: Don’t hardcode today’s data
Next, identify:
what will differ each execution.
Examples include:
- Project
- Board
- Sprint
- Date range
- Team
- This week’s priorities
- Special events
Don’t lock these as permanent rules.
For example, today you might ask:
“Summarize Project Alpha’s Sprint 23.”
If you save:
Project Alpha
and
Sprint 23
inside the Skill,
next week you’ll still be using the old scope.
A better way:
Keep the Skill fixed to:
“Please summarize the specified Board and date range.”
Then provide at runtime:
“Which Board this time?”
“What date range?”
This means:
Fixed method, variable data.
③ Mandatory Human Confirmation: Don’t embed high-impact actions in the fixed process
The third layer is crucial.
Suppose a Skill doesn’t just create a summary, but also:
- Publishes a Confluence page
- Changes Jira status
- Sends external emails
- Deletes work items
- Makes customer commitments
- Alters official deadlines
Don’t assume:
“Because this happens weekly,”
it means
“It can be fully automated indefinitely.”
Repetition and low risk
are not the same.
For example:
Weekly status updates must be sent.
“Weekly sending”
is a fixed process.
But:
Whether the email content is ready to send,
may still require human review.
So design the Skill:
Organize data
→ produce a draft
→ highlight missing info
→ pause for human review
instead of:
Organize
→ send out.
Understanding by example
Imagine you want to create a:
Weekly Leadership Update Skill.
Don’t immediately tell Rovo:
“Save my current prompt.”
Break it into three parts first.
Fixed Rules
Always include:
- This week’s completions
- Main blockers
- Next week’s priorities
- Issues needing leadership decisions
- No conclusions without supporting data
Variable Inputs
Provide each time:
- Which project
- Date range
- Which team
- Special events this week
- Whether there is a specified audience
Mandatory Human Confirmation
After AI finishes,
only generate a draft.
Matters involving:
- New deadlines
- Budget commitments
- Customer promises
- Personnel responsibilities
must be approved by leadership first.
This approach better mirrors a truly repeatable workflow.
Why is this especially important for Rovo?
Atlassian’s newest Custom Skills aren’t just about saving a text snippet.
The official example is:
You can directly tell Rovo:
Create a Skill to generate Sprint Summaries from a Board.
Rovo will ask follow-up questions,
understanding:
- Edge cases
- Expected outputs
- Work methods
and then organize everything into a reusable Skill.
So when building a Skill,
the key is not:
“How pretty is the prompt?”
but rather:
“What should stay fixed, and what must be updated each time?”
Skills can also execute tasks
Rovo Skills are more than text templates.
According to Atlassian,
Skills let Rovo act on your behalf inside the app.
Examples include:
- Creating work items
- Updating work items
- Generating status updates
- Publishing content
- Handling other work actions
Here, mandatory human confirmation becomes even more important.
Atlassian notes that Skills usually prompt users for confirmation when performing actions,
but Agents inside Automation Rules might behave differently.
So don’t treat:
“The platform might have confirmation”
as
“Your workflow doesn’t need its own human approval step.”
The platform’s safety controls are one layer.
Workflow decisions are another.
One final way to decide
Before turning repetitive AI work into a Skill,
ask about every rule three times:
“Will this still apply next week?”
If yes:
→ Fixed rule
“Will this change next week?”
If yes:
→ Variable input
“If this is done wrong, will it really affect others, data, or commitments?”
If yes:
→ Mandatory human confirmation
Only after sorting these three categories,
then build your Custom Skill.
Don’t do it the other way:
Save the entire prompt,
and after it fails, figure out which parts shouldn’t have been fixed.
A sign of mature AI workflows
isn’t:
“I can finally automate this.”
But rather:
“We’ve clarified what stays fixed, what updates each time, and where to pause for human review.”
If you want help identifying the best part of your process to hand off to AI, comment “workflow.”
Today, let’s improve with AI a little.
Learn one AI skill daily.
Save a little time every day.
Boost your skills a bit each day.
SasaDaily, growing with you.
