Skip to content

sprint-spine.md

Latest commit

 

History

History
143 lines (107 loc) · 6.56 KB

sprint-spine.md

File metadata and controls

143 lines (107 loc) · 6.56 KB

BA-Kit Sprint Spine

The unified 7-phase loop that all 44 BA-Kit agents feed into. Think→Plan→Build→Review→Test→Ship→Reflect — adapted for Requirements Engineering.

Why a spine?

BA-Kit v3.1 had 33 agents + 17 cookbook scenarios. v3.4 added 11 gstack-distilled sprint spine agents, and v3.5 ships 44 total (33 original with anti-rationalization pattern + 11 sprint spine). That's powerful, but the user faces paralysis of choice: "Which agent first? Which scenario fits?"

Gstack (the reference for this design) solves this with a single sprint loop that every skill feeds into. One loop, many nodes.

This doc defines the equivalent for BA: the Requirements Sprint Spine.

The 7 phases

Discover → Elicit → Define → Validate → Prioritize → Publish → Reflect
                                                                    │
                                                                    └──► (loop back)
Phase Question the phase answers Output
1. Discover "What's the business context? Who cares?" Stakeholder map, strategic canvas, competency grid
2. Elicit "What do stakeholders actually need?" Interview notes, assumption list, raw backlog
3. Define "What are the precise requirements?" User stories, AC, NFRs, data models, business rules
4. Validate "Are the requirements correct, complete, testable?" Quality gate report, consistency report, test cases
5. Prioritize "What ships first? What's the trade-off?" MoSCoW/RICE/WSJF ranking, prioritized backlog
6. Publish "How does this reach the team + systems?" Jira tickets, Confluence pages, DOCX exports
7. Reflect "What did we learn? What drifted?" Retro, as-built drift, captured learnings

Agent → phase map

Phase Primary agents Supporting agents
1. Discover @ba-identity, @ba-strategy, @ba-systems @ba-questioning, @ba-ux
2. Elicit @ba-elicitation, @ba-questioning, @ba-facilitation @ba-wiki (prior art), @ba-ux
3. Define @ba-writing, @ba-nfr, @ba-data, @ba-business-rules, @ba-process @ba-diagram, @ba-ux
4. Validate @ba-validation, @ba-consistency, @ba-quality-gate, @ba-test-gen, @ba-traceability @ba-challenger 🆕, @ba-second-opinion 🆕
5. Prioritize @ba-prioritization, @ba-solution, @ba-conflict @ba-agile, @ba-innovation
6. Publish @ba-export, @ba-jira, @ba-confluence, @ba-communication @ba-change
7. Reflect @ba-retro 🆕, @ba-as-built 🆕, @ba-auditor, @ba-metrics @ba-learn 🆕, @ba-root-cause

Meta-agents (span all phases):

  • @ba-master — dispatcher
  • @ba-autoreview 🆕 — runs phase 4 pipeline in one command
  • @ba-checkpoint 🆕 — session save/resume (works at any phase)
  • @ba-wiki — knowledge query (used anywhere)

How to move between phases

Each phase has an exit criterion. Don't advance without satisfying it.

From → To Exit criterion
Discover → Elicit Stakeholder map signed off, top 3 strategic drivers identified
Elicit → Define ≥ 3 raw requirements captured per module; no "TBD" blockers
Define → Validate Every requirement has acceptance criteria + owner
Validate → Prioritize Quality gate PASS or CONDITIONAL (not REJECT); consistency clean
Prioritize → Publish Ranking agreed by PM + stakeholders; dependencies mapped
Publish → Reflect Tickets + docs delivered to downstream systems
Reflect → Discover Retro action items feed next discovery loop

Hand-off contract (SKILL.md frontmatter v2)

Starting v3.4, every SKILL.md should declare its phase + data contract:

---
name: ba-writing
description: ...
version: 1.1.0
phase: Define                    # 1 of 7 phases above
inputs:                          # What must exist before this skill runs
  - elicitation-notes.md
  - stakeholder-map.md
outputs:                         # What this skill produces
  - user-stories.md
  - acceptance-criteria.md
downstream:                      # Which skills consume my output
  - ba-validation
  - ba-test-gen
  - ba-consistency
---

Chains form automatically: if you run @ba-writing and its downstream field lists @ba-validation, the dispatcher (ba-master) can auto-suggest the next step.

Consolidating the 17 cookbook scenarios

The existing docs/workflow-cookbook.md has 17 scenarios. They are not deleted — they become variants of the spine:

Scenario Spine entry point Special variant
#1 "Startup Founder" Discover → ... → Publish Full loop, MoSCoW heavy
#2 "Legacy Migration" Discover (as-is) → Define → Validate Adds @ba-process at Discover
#3 "Stakeholder War" Elicit → Prioritize (conflict branch) @ba-conflict at Prioritize
#5 "Production Crisis" Reflect → Define Root cause loop
#8 "Compliance Audit" Validate → Publish @ba-traceability heavy
#17 "Testing Handoff" Validate → Publish @ba-test-gen + @ba-traceability
... ... ...

Rule: Every cookbook scenario must map to a contiguous path through the spine. If it doesn't, it's flagged as "scenario debt" and either normalized or removed.

Running the spine as a single command

The new @ba-autoreview skill runs Phase 4 (Validate) end-to-end:

@ba-autoreview outputs/my-project/
  → Phase 0: Detect artifact types (BRD / US / API / DB)
  → Phase 4a: @ba-consistency    (cross-artifact alignment)
  → Phase 4b: @ba-quality-gate   (8-dim scoring)
  → Phase 4c: @ba-traceability   (RTM coverage)
  → Phase 4d: @ba-auditor        (synthesize health report)
  → (optional) @ba-second-opinion (cross-model review)
  → Single verdict: PASS / CONDITIONAL / REJECT

Future work: ba-autodiscover, ba-autoelicit (names reserved, not yet implemented) for other phases.

Versioning

  • v1.0 (2026-04-13): Define 7 phases + initial agent mapping.
  • v1.1 (v3.4, 2026-04-13): Added 11 sprint spine agents via gstack distillation.
  • v1.2 (v3.5, 2026-04-14): Reconciled with v3.2/v3.3 anti-rationalization rollout — 44 total agents, 33 with anti-rationalization pattern + 11 sprint spine (anti-rat inheritance TBD).
  • v2.0 (future): Runtime enforcement — ba-master blocks out-of-phase handoffs + frontmatter phase: on all 44 skills.