agentic-sdlc
End-to-end agentic SDLC: brief → blast radius → ARB → bolt → compliant PR
This plugin encodes a production-grade agentic software development lifecycle based on the BCG Dark Factory model, Instructure's 90%+ Claude Code adoption experience, and Fortune 500 best practices as of April 2026. It gives you a structured pipeline where a work item enters as a brief and exits as a compliant, tested pull request — with human oversight at every critical gate.
The Model
Bolts replace sprints. A bolt is a compressed delivery unit: brief-in to PR-merged, measured in hours or days. The orchestrator dispatches specialist agents in parallel, enforces compliance gates unconditionally, convenes the Architecture Review Board for High+ blast radius changes, and commits institutional knowledge back to the repository after every bolt.
Brief → Blast Radius → [ARB] → [Human Gate] → Parallel Agents → Quality Gate → PR → Memory Write
Commands
| Command | What it does |
|---|
/brief | Socratic dialogue to produce a brief-schema.yaml |
/bolt | Execute the full pipeline from brief to PR |
/arb | Convene a manual Architecture Review Board session |
/topology | Initialize or view the project topology.yaml |
Skills
| Skill | Triggers on |
|---|
sdlc-orchestration | Bolt execution, phase guidance, compliance architecture, audit log schema |
git-standards | Branch naming, commit format, PR structure, merge strategy |
memory-system | Writing/reading .claude/memory/, decision records, lessons, standards registry |
Agents
| Agent | Role | Model |
|---|
orchestrator | End-to-end pipeline coordination | Opus |
arb-architect | System design, standards compliance, Round 3 chair | Sonnet |
arb-senior-engineer | Implementation feasibility, tech debt | Sonnet |
arb-data-engineer | Data model, migration safety, pipelines | Sonnet |
arb-infosec | Threat model, auth, secrets, compliance | Sonnet |
arb-sre | Observability, rollback, SLO impact | Sonnet |
arb-qa | Testability, acceptance criteria, regression risk | Sonnet |
Hooks
| Hook | Event | What it does |
|---|
pre-write-check.sh | PreToolUse (Write/Edit) | Blocks secrets file writes; validates branch naming |
pre-commit-check.sh | PreToolUse (Bash) | Enforces ticket reference in git commit messages |
| PostToolUse prompt | PostToolUse (Write/Edit) | Secondary secrets scan on written content |
session-start.sh | SessionStart | Surfaces memory index, topology, and active brief |
Setup
1. Initialize your project topology
/topology init
This creates topology.yaml — the single file that tells every agent which tools you're using (GitHub vs. GitLab, Jira vs. Linear, etc.).
2. Create your first brief
/brief
The agent conducts a Socratic dialogue and writes brief-schema.yaml.
3. Run the bolt
/bolt
The orchestrator takes it from there: blast radius, ARB (if triggered), implementation, tests, PR.
Toolchain Support
This plugin is toolchain-agnostic. See CONNECTORS.md for the full list of supported tools. Configure your toolchain by running /topology init — the plugin adapts to whatever you declare there.
Compliance
The pipeline enforces SOC 1/2, ISO 27001, NIST AI RMF, EU AI Act, and HIPAA-compatible controls by default:
- Immutable audit trail (append-only JSONL)
- Secrets blocked unconditionally by PreToolUse hook
- PII guardrails in PostToolUse hook
- Human gates at every architecture decision and production promotion
- Blast radius scored before any code generation
- ARB convened automatically for High+ blast radius and compliance-flagged work
The Architecture Review Board
When the ARB is triggered, six specialist agents analyze the design in parallel across three rounds:
Round 1 — Independent analysis (no cross-panelist visibility)
Round 2 — Deliberation (CORROBORATE / CHALLENGE / ESCALATE / AMPLIFY / WITHDRAW)
Round 3 — Architect chairs synthesis, produces Panel Synthesis for human decision
Corroboration weighting determines final severity: unanimous findings are non-waivable; supermajority findings are Critical; solo InfoSec/Architect primary domain findings are High.
The human makes the final call. Always.
Self-Evolving Memory
After every bolt, agents commit institutional knowledge to .claude/memory/:
decisions/ — architecture decisions with rationale and alternativeslessons/ — errors, corrections, rollbacks (with prevention rules)patterns/ — recurring implementation patternsstandards/ — ARB-derived architectural rules (enforced on all future designs)
Memory accumulates across bolts. The system gets measurably better over time.
xrf9268-hue
JuliusBrussee
wshobson
alirezarezvani
