BA-Kit v3.5.0 is an agent squad of 44 specialists for Requirements Engineering. Multi-platform: Antigravity IDE, Claude Code, Claude CoWork.
Standards: BABOK v3, IEEE 29148, ISO 25010, INVEST.
.agent/skills/ba-*/SKILL.md — 44 Agent Skill definitions + 2 connectors + 1 meta
.agent/skills/_shared/ — Shared utilities & knowledge search skill
.agent/scripts/ — BM25+ search, ba_e2e_test orchestrator, Python helpers
.agent/data/*.csv — 831 indexed knowledge entries (23 domains)
.agent/templates/ — 14 BA document templates (BRD, SRS, FRD, etc.)
.agent/wiki/ — Living knowledge wiki (2-tier: concepts/decisions)
.github/workflows/ — CI: e2e-skills.yml (Phase 08 quality gate)
docs/ — Documentation and guides (incl. sprint-spine.md)
outputs/mini-app-cham-cong/ — BA showcase reference (12 modules, 53 US)
ebooks/ — Synthesized book knowledge
# Knowledge search
python3 .agent/scripts/ba_search.py "query" --domain elicitation
python3 .agent/scripts/ba_search.py "query" --multi-domain
# Coverage checker
python3 .agent/scripts/coverage_checker.py outputs/project-name/
# Generate DOCX output
python3 .agent/scripts/gen_docx.py input.md output.docxAgents are activated via @ba-{name} in Antigravity or as SKILL.md files in Claude Code.
@ba-master is the dispatcher — route all unclear requests through it first.
@ba-master— start here for any new request@ba-elicitation— gather requirements from stakeholders@ba-writing— write US/BRD/SRS/API specs@ba-validation— INVEST check, quality review@ba-quality-gate— PASS/CONDITIONAL/REJECT scoring (8 dimensions)@ba-auditor— full project health dashboard
Every agent follows System 2 Reflection: draft → critique → refine.
- ALWAYS verify math with Python, links with Grep, standards with WebSearch
- NEVER assume user intent — use
@ba-elicitationto clarify - NEVER hallucinate file contents — check with Grep/Read tools
Agents hand off via structured output blocks. See .agent/skills/_shared/ for
shared handoff templates and cross-agent conventions.