lucavb Jira ticket refinement
Help the user refine stories so engineers can pick them up. Read the ticket, trace the codebase, surface plot holes and PO gaps, draft a Jira comment, fix typos on the ticket directly. Goal: the PO learns to write better stories.
All Jira reads and writes go through Atlassian MCP (plugin-atlassian-atlassian). Read tool schemas before calling.
Refinement loop (one ticket per pass)
Work tickets in the order the user gives. Do not batch-post comments.
1. Ingest
- Fetch the ticket:
getJiraIssuewithresponseContentFormat: "markdown". Pull comments and linked issues when sizing cross-ticket conflicts. - Fetch sibling tickets in the same epic when the user names a set or overlap is likely.
Done when: You can state the story, user journey, AC, and out-of-scope in plain language.
2. Legwork
Apply discover-before-planning: resolve what you can from the codebase before briefing the user.
Trace the ticket against the codebase before critiquing. Spawn explore agents in parallel when the epic spans multiple areas.
Order: read what the ticket says → trace code → flag engineering gaps for the user → surface PO questions only when ticket + code still leave genuine product ambiguity.
Done when: You know what exists today, what the ticket assumes, and which gaps are PO-owned vs engineering-owned.
3. Brief the user
Return a concise summary:
- What the ticket is about (2-4 sentences)
- What is clear and good
- Plot holes - AC gaps, contradictions with other tickets, Figma vs ticket mismatches, implementation leaking into requirements, ambiguous journeys
- Engineering notes for the user only (repo paths, missing APIs) - not for the PO comment
See po-critique.md for what belongs in PO critique vs what to keep out.
Done when: Every plot hole maps to a PO decision or a missing AC, not an implementation preference.
4. Draft the PO comment
Write the comment per comment-voice.md. Show the full draft to the user. Do not post until they approve.
Done when: Draft is shown and you are waiting for post / edit / skip.
5. Post and patch
On approval:
- Post via
addCommentToJiraIssuewithcontentFormat: "markdown". - Fix obvious typos on the ticket directly (spelling, grammar, copy mismatches with Figma). Do not comment on typos you fixed.
For ticket description edits on story tables, follow the atlassian-jira-adf-formatting skill: ADF round-trip (getJiraIssue → edit → editJiraIssue with contentFormat: "adf"). Validate with node ~/.cursor/skills/atlassian-jira-adf-formatting/scripts/validate-adf.js. Re-fetch to confirm structure.
Large ADF payloads: write the full editJiraIssue arguments to a JSON file, pass the entire parsed object to MCP. If inline args fail, use a subagent to call MCP with the file. Never send a partial ADF - that truncates the description.
Done when: Comment is posted (or user skipped) and any ticket typo fixes are verified via getJiraIssue.
Then ask whether to continue to the next ticket.
Branches
| Branch | When | Skill |
|---|---|---|
| Ticket typo / copy fix | Spelling, grammar, Figma copy drift | atlassian-jira-adf-formatting |
| Figma check | Design is source of truth for epic UX | User-Figma MCP or user checks manually |
| Epic-wide pass | User names multiple keys | Run the loop per ticket; note cross-ticket conflicts in each brief |
Cloud ID
Resolve once per session via getAccessibleAtlassianResources, or reuse a known cloudId from a ticket already fetched in the conversation.