jira-intelligence
Agent-native equivalents of Jira Intelligence (Atlassian AI) — no Premium license required. Seven commands: work breakdown (epic → child stories), story enhancement (rewrite vague tickets), comment th
SKILL.mdskills/cross-cutting/jira-intelligence/SKILL.md
GitHubJira Intelligence (Agent-Native)
Context
Atlassian Intelligence is a Jira Premium/Enterprise feature. This skill replicates its core capabilities using the AI agent directly — paste any ticket, epic, comment thread, or natural language query and get the same output. Works with plain text input or Jira MCP when configured.
Seven commands — invoke any independently:
| Command | Trigger phrase |
|---|---|
| Work Breakdown | "Break this epic into stories" |
| Story Enhancement | "Improve this ticket" |
| Comment Summary | "Summarize this ticket" |
| Edge Case Generator | "What are the edge cases?" |
| Dependency Detector | "Find dependencies in these stories" |
| Story Point Estimator | "Estimate this story" |
| NL → JQL | "Turn this into JQL" |
1. Work Breakdown
"Break this epic into stories" / "Generate child issues from this epic"
Input: Paste the epic title + description — or fetch via MCP:
jira_get_issue(issue_key: "PROJ-epic")
What the agent does:
- Identifies core user journeys and outcomes in the epic
- Applies vertical slicing patterns (workflow steps, happy/unhappy paths, CRUD, roles)
- Checks each candidate against INVEST (Independent, Valuable, Estimable, Small, Testable)
- Returns child stories with title, acceptance criteria, size signal, and splitting rationale
Output:
Epic: <title>
Split strategy: <pattern used — e.g. Workflow Steps + Happy/Unhappy Path>
Child Stories:
1. [S] <title>
Value: <one-line user benefit>
AC:
- Given <ctx>, When <action>, Then <outcome>
Splitting rationale: <why this is its own story>
INVEST: I✅ N✅ V✅ E✅ S✅ T✅
2. [M] <title>
...
Create in Jira (MCP):
jira_create_issue(project: "PROJ", issue_type: "Story", summary: "<title>",
description: "<AC + rationale>", parent: "PROJ-epic", labels: ["vertical-slice"])
Tip: Combine with the
story-slicingskill for codebase-grounded sizing.
2. Story Enhancement
"Improve this ticket" / "Rewrite this story" / "This description is vague"
Input: Paste the raw ticket title + description in any format.
What the agent does:
- Rewrites using
As a / I want / So thattemplate - Fills in or completes acceptance criteria as Given/When/Then
- Flags INVEST violations with specific fixes
- Adjusts tone for the right audience (engineering, product, or stakeholder)
- Ensures scope is bounded (not over-specified, not under-specified)
Example:
Before: "Add search to the app"
After:
Title: User can search products by keyword
As a shopper
I want to search for products by entering a keyword
So that I can find what I need without browsing every category
Acceptance Criteria:
- Given I type 3+ characters, When I submit search, Then results matching name or
description appear ranked by relevance
- Given no results match, Then "No results for X" message shown with a reset link
- Given I clear the search field, Then the full product list is restored
INVEST fixes applied:
❌ Not Estimable — scope was unbounded → ✅ Scoped: keyword match on name + description only, no fuzzy
❌ Not Testable — no AC → ✅ 3 Given/When/Then criteria added
Update in Jira (MCP):
jira_update_issue(issue_key: "PROJ-123", fields: {
summary: "<enhanced title>",
description: "<enhanced description with AC>"
})
3. Comment Thread Summary
"Summarize this ticket" / "TLDR this thread" / "What's the current status?"
Input: Paste the comment thread — or fetch via MCP:
jira_get_issue(issue_key: "PROJ-123") # includes description + all comments
Output:
## Summary: PROJ-123 — <issue title>
Status: <current status> Last updated: <date>
### What's been decided
- <decision 1 with date/author>
- <decision 2>
### Key context
- <important constraint or background>
### Open questions
- <unresolved item> — needs input from <person/team>
### Action items
- [ ] <action> — owner: <name>
- [ ] <action> — owner: <name>
### What needs to happen next
<one-paragraph recommendation>
4. Edge Case Generator
"What are the edge cases for this story?" / "What am I missing in the AC?"
Input: Paste the story title + acceptance criteria.
Output:
Story: <title>
Happy path covered: ✅
Edge cases by category:
🔴 Error / failure states
- Empty input submitted
- Input exceeds max length (what's the limit?)
- Upstream service returns 500
🟡 Boundary conditions
- Exactly 0 items vs. 1 vs. the maximum
- Date is today / yesterday / tomorrow / far future
🟠 Authorization
- Non-owner attempts access
- Session expires mid-flow
- Role lacks required permission
⚪ Empty / null states
- No results match the query
- Required field is null in the database
- First-time user with no history
Suggested additional AC (ranked by risk):
1. [HIGH] Given empty input submitted, Then inline validation error shown
2. [MEDIUM] Given session expires during the flow, Then redirect to login with state preserved
3. [LOW] Given 0 results match, Then "No results" message with reset action shown
5. Dependency Detector
"Find dependencies in these stories" / "Which stories are blocking each other?"
Input: Paste a list of stories, or fetch the sprint via MCP:
jira_search_issues(jql: "sprint in openSprints() AND project = PROJ ORDER BY rank ASC")
What the agent does:
- Identifies shared data entities, API endpoints, auth preconditions, and sequential user flows
- Classifies as hard dependency (blocked) or soft dependency (integration risk)
- Suggests a delivery sequencing order
Output:
Dependency Analysis — <Sprint or Epic>
⛔ Hard dependencies (cannot start until blocker is done):
PROJ-12 → BLOCKED BY → PROJ-8 (shares Order entity)
PROJ-15 → BLOCKED BY → PROJ-12 (admin view requires checkout to exist)
⚠️ Soft dependencies (can start in parallel; integration needed):
PROJ-10 and PROJ-12 — both write to the Order table
PROJ-11 and PROJ-9 — both use the same Stripe credentials
✅ Fully independent:
PROJ-13 PROJ-14
Suggested sequencing:
Week 1: PROJ-8 → PROJ-9
Week 2: PROJ-10, PROJ-11, PROJ-12 (parallel)
Week 3: PROJ-13, PROJ-14, PROJ-15
Create dependency links in Jira (MCP):
jira_create_issue_link(inward_issue: "PROJ-12", outward_issue: "PROJ-8", link_type: "is blocked by")
6. Story Point Estimator
"Estimate this story" / "How complex is this?" / "Give me a point estimate with reasoning"
Input: Story title + description. Optionally, paste codebase scan output for higher accuracy.
Estimation model (Fibonacci: 1, 2, 3, 5, 8, 13):
| Complexity signal | Points |
|---|---|
| Base (any story) | 1 |
| Each additional service touched | +1 |
| New DB table or migration | +2 |
| New external integration / third-party API | +3 |
| Auth / permission change | +2 |
| > 5 acceptance criteria | +1 |
| Team unfamiliar with this area of codebase | +2 |
| Cross-platform (web + mobile) | +2 |
| Performance or scalability NFR | +2 |
Output:
Story: <title>
Estimate: 5 points
Reasoning:
Base → 1
1 new endpoint (POST /sessions) → +1
JWT auth token generation → +2
Team hasn't touched auth in 6mo → +2 (unfamiliarity)
Total: 6 → rounded to 5 (Fibonacci)
Confidence: Medium
If confidence is low: consider a 1-point spike first.
Spike focus: <specific unknown to investigate>
7. Natural Language → JQL
"Find all stories stuck in progress for more than 5 days" "Show me unresolved bugs assigned to me this sprint"
What the agent does: Translates plain English into valid JQL, then optionally executes it via MCP.
Examples:
| Plain English | JQL |
|---|---|
| Stories in progress for more than 5 days | project = PROJ AND status = "In Progress" AND status changed to "In Progress" before -5d |
| Unresolved bugs assigned to me this sprint | project = PROJ AND issuetype = Bug AND status != Done AND assignee = currentUser() AND sprint in openSprints() |
| Epics with no child stories | project = PROJ AND issuetype = Epic AND issueFunction not in subtasksOf("issuetype = Story") |
| Stories completed last sprint | project = PROJ AND sprint in closedSprints() AND status = Done AND issuetype = Story |
| High-priority blockers from last 2 weeks | project = PROJ AND labels = blocked AND priority in (High, Highest) AND created >= -14d ORDER BY created DESC |
| Stories without an assignee in this sprint | project = PROJ AND sprint in openSprints() AND assignee is EMPTY |
| All stories that changed status today | project = PROJ AND status changed after startOfDay() |
Execute via MCP:
jira_search_issues(jql: "<generated JQL>", max_results: 25)
Graceful fallback: Output the JQL as a copyable string — paste directly into Jira's issue navigator.
Agent Instructions
When applying this skill, the agent should:
- Identify which command the user is invoking from their trigger phrase
- Ask for the input (ticket text, epic, comment thread) if not provided — or fetch via MCP
- Apply the relevant command and produce the structured output
- If Jira MCP is available, offer to execute the result (create tickets, update fields, run JQL)
- If multiple commands are useful (e.g., enhance a story and generate edge cases), chain them automatically
MCP Quickref
jira_get_issue(issue_key: "PROJ-123")
jira_search_issues(jql: "<jql>", max_results: 25)
jira_create_issue(project: "PROJ", issue_type: "Story", summary: "...", description: "...", parent: "PROJ-epic")
jira_update_issue(issue_key: "PROJ-123", fields: { summary: "...", description: "..." })
jira_create_issue_link(inward_issue: "PROJ-A", outward_issue: "PROJ-B", link_type: "is blocked by")
jira_transition_issue(issue_key: "PROJ-123", transition: "In Progress")
References
- Atlassian: AI Features in Jira
- Bill Wake: INVEST in Good Stories
- Richard Lawrence: Story Splitting Flowchart
- LeadDev: Making refinement less painful