Add knowledge from jira

add-knowledge-from-jira (Jira → business-rule evidence)

Pipeline: Ingest → Classify → Recognize (shared confirm with wiki) → Compose (unified S3→S7). AC, comments, and last-wins decisions are first-class domain evidence — they enter the S3/S4/S5 mainline, not a post-brief patch.

confirm authorizes Compose; it is not “briefs already done”. (zh-CN strings for these tokens: domain-knowledge/language/deliverable-locale-tokens.json.)

Compose language (same as wiki): S6 = source-language brief; S7 = convert to deliverable_locale (expression only).

When to use

• Fold sprint/history ticket rules, AC, and comment decisions into the same domain library
• Not for single-ticket risk/split/test-design → @requirement-risk / @ticket-splitter / @ticket-test-design

Prep vs Compose (stop boundaries)

Label	Contains	Who	Stops at
Script prep	Ingest + Classify	run_jira_knowledge.py (no flag = board history; --sprint-id = one sprint)	After domain_check jira --full-raw
Full prep	Script prep + Cursor Recognize	Agent (wiki RUNBOOK · S2)	Checklist + closure ready → wait for confirm
Compose	Unified S3→S7	Agent + wiki RUNBOOK	Only confirmed themes; S3 includes eligible Jira materializations

User mode= → CLI

User mode=	Agent runs
history + team=<t>	python3 scripts/run_jira_knowledge.py --team <t> (board history through current sprint) + Classify
history + board-id=<id>	Resolve team via team-roots.json, then same as history
sprint + sprint-id=<id>	run_jira_knowledge.py --team <resolved> --sprint-id <id> + Classify
(default / prep)	history if team/board given; sprint if sprint-id given
compose / continue	Wiki RUNBOOK Compose (S3→S7); S3 reads Confluence and Jira evidence
distill / reconcile / full	Removed — refuse; use prep → Recognize → Compose

Hard rules

1. Ingest (script): jira/raw/ — INGEST-CLI.md
2. Classify (script): attribution/ + theme indexes — CLASSIFY-CLI.md
3. Ticket count ≠ done: Classify coverage is not pipeline completion (pipeline-design.md)
4. Jira is not an appendix: distill_tier=proposition_core/platform_variant tickets enter S3 proposition candidates
5. Recognize shares DOMAIN_MODULE_CHECKLIST.md with wiki; confirm authorizes Compose
6. Compose follows generate-knowledge-from-wiki/RUNBOOK.md (S6 source brief → S7 locale brief)
7. Legacy Extract/Integrate paths are deleted — do not revive post-S7 Jira patches
8. primary / themes[] = confirmed slugs; gateway / requirements are sink buckets, not domains
9. Source of truth is Jira; do not treat hand-edited jira/materialized/ as authority
10. No in-repo HTTP LLM API
11. No translation in S1–S6; locale conversion only in S7

User examples

@add-knowledge-from-jira team=demo mode=history
@add-knowledge-from-jira board-id=1 mode=history
@add-knowledge-from-jira team=demo mode=sprint sprint-id=1726

Review checklist → confirm → continue → verify aggregates/work drafts/S6+S7 briefs include Jira rules.

Agent checklist

#	Action
1	Ingest: run_jira_ingest.py or run_jira_knowledge.py (INGEST-CLI.md)
2	Read JIRA_PIPELINE_HANDOFF.json → curated/by-root/<R>/
3	Classify (CLASSIFY-CLI.md) → Recognize → pause at full prep
4	On continue (confirm rows only): Compose S3→S7; proposition lists should show source_system: jira
5	After Compose: domain_check distill --root-id <R>; optionally domain_check jira --full-raw

Forbidden

• Report “pipeline done” because N tickets have attribution
• Claim Jira rules merged from the gap-scan index (by-theme/<team>/ output of read_business.py) alone
• Compose without confirm
• Promote gateway / requirements to confirmed checklist rows
• Run deleted legacy scripts or post-S7 patches
• Skip S6 and write only a translated S7 file

Further reading

Doc	Purpose
RUNBOOK.md	Jira stages + shared layers
RUNBOOK.zh-CN.md	Chinese locale playbook
INGEST-CLI.md / CLASSIFY-CLI.md	Script flags & gates
generate-knowledge-from-wiki/RUNBOOK.md	Shared Recognize + Compose
../../contracts/domain-knowledge-pipeline-contract.md	Artifact layers

Next

@requirement-risk → @ticket-splitter → @ticket-test-design (read S7 Compose briefs).