Ticket splitter

ticket-splitter (INVEST backlog slices)

First principles

Artifact: iteration-ready commitments — each item has a boundary (scope) and an observable completion face (done_when). Not a task list or “dev vs QA” role split.

Quality: done_when must say what done looks like and how it is seen (user behavior / system-observable state / contract). “Dev done”, “ready for QA” are not completion faces. Title = 3-second headline (one primary outcome). scope bullets = one idea each (P11/P12).

	Wiki @generate-knowledge-from-wiki	This skill
Cost of error	Pollutes library briefs	Planning meeting rewrites a few items
Who stops bad output	Human confirms modules	Scripts block fake testability / missing structure
Loop	Prep → confirm → compose → gate	Draft → gate → fix failures only → gate

Corollaries

1. Do not wait for human confirm in the loop.
2. Borrow wiki’s script-verified explicit form, not confirm/continue ritual.
3. Scripts do not decide “how many slices”.
4. Substance: ticket_system.md; presentation: references/presentation.md.

When to use

• Planning: cut one Jira story into executable backlog items
• Need observable scope / done_when
• Optional domain briefs under curated/by-root/<root_id>/

When not

Scenario	Use
Risk only	@requirement-risk
Refresh wiki / write briefs	@generate-knowledge-from-wiki / @distill-domain-knowledge
Sprint bulk into briefs	@add-knowledge-from-jira
Requirement unreadable	State blockers; do not interrogate endlessly

How to invoke

• @ticket-splitter DEMO-1 team=demo (offline demo)
• @ticket-splitter PROJ-123
• @ticket-splitter team=<key> PROJ-123
• @ticket-splitter PROJ-123 + code paths (R2.5 anchors)
• @ticket-splitter PROJ-123 brief
• Or pasted requirement text

Offline DEMO-*: ../_shared/offline-demo.md.

Param	Values	Default
team	key/alias in team-roots.json	attribution / Agile Team
brief	short-mode words in message	full items

Preconditions & failures

• Offline DEMO-*: no credentials
• Real keys: .env Atlassian/Jira credentials + network; fetch_jira_attachments.py as needed
• No credentials → paste text or DEMO-1
• No brief → still emit; note missing anchors
• Gate fail → fix and re-run; do not claim done without pass

Decision tree

Jira key?
├─ DEMO-* / offline → fixtures/offline-demo; skip network
├─ real key → attachments via MCP or fetch script
├─ Resolve team/root
├─ primary slug → S7 locale brief (`*-domain-brief.md` / `*-领域知识定稿.md` per locale); else S6 source-brief / work draft (noted)
├─ Jira says “dev only, test elsewhere”? → do not invent test rows; note in narrative
├─ brief mode? → Scope + split overview + correction note (still gated)
├─ Draft → .jira_attachments/<KEY>/split_draft.md when KEY present
└─ Gate: validate_ticket_split.py [--brief] → fix → deliver

Responsibility

Layer	Agent	Script
Evidence	Which briefs / code anchors	Fetch issue & attachments
Generate	Split + presentation	No prose authorship
Gate	Fix from report	validate_ticket_split.py
Semantic review	Only if user asks for principles	None

Gate rules

1. Full and brief must pass validator before delivery.
2. Fake testability (done_when = “dev done / ready for QA” with no observable state) → fail.
3. Need Scope + ## Split overview (or locale equivalent the validator accepts); full items need title/scope/done_when/deps/confidence.
4. P11 title anti-patterns and P12 welded scope bullets → fail when detectable.
5. On fail, fix only reported issues.
6. Standards: ticket_system.md + presentation; scripts verify only.
7. Never write curated/ or run distill check.

Presentation & domain library

• references/presentation.md · P10: ../_shared/presentation-p10.md
• ../../contracts/jira-issue-domain-knowledge-context.md
• Anchor scope / done_when to S7 locale briefs when present (R2.5); if only S6 / work draft exists, note “not S7” and still constrain to that evidence.

Unlike @add-knowledge-from-jira, this skill outputs iteration backlog items, not library merges.

Agent checklist

1. Load Jira + domain evidence; fetch attachments if needed.
2. Split per ticket_system; constrain with briefs; write correction narrative when needed.
3. Presentation + P11/P12 self-check; attach risk R-00N when same-session risk exists (P7).
4. validate_ticket_split.py; fix until pass; deliver in chat.
5. Do not default-wait for human confirm.

Done when

• Validator exit 0 (or explicitly marked failed)
• Scope + split overview present
• Full items have observable done_when; no fake-testability ban words
• Titles/scope pass P11/P12 intent
• Risk R-00N linked when applicable
• Not split primarily as “dev vs test” for the same outcome

Forbidden

• Write curated/ / run distill check
• Items that contradict explicit brief out-of-scope
• Invent subsystem names absent from briefs
• Skip gate or wait for confirm unless ungated draft requested
• Copy wiki confirm→continue as the split main flow
• Weld multi-concern scope into one bullet; empty verbs as fake coverage

Handoff to @ticket-test-design

After a usable split (or instead of split when AC are already clear), testers may @ticket-test-design <same KEY> to produce must/should/later cases proving AC / done_when.

Examples

@ticket-splitter DEMO-1 team=demo
→ offline → draft → validate → deliver

@requirement-risk PROJ-1 then @ticket-splitter PROJ-1
→ no confirm wait; link R-00N; note open MUST assumptions