Ship Command v2.0 - Intelligent Orchestration

End-to-end workflow: PREFLIGHT → QUESTIONS → PREP → EXPLORE/PLAN/CODE/TEST/FIX/DOCUMENT → COMMIT/PR → CI → COUNCIL → SHIPPED

Features: Question-first mode, state machine with resume, pre-flight validation, dry-run mode, CI integration, dynamic council selection

---

Phase 0: Pre-Flight Validation

Validate all requirements before work begins:

Checks:

• Jira: API connection, issue exists, not Done/Closed, user has permissions
• Git: in repository, working directory clean, can push to remote
• Harness/CI: API configured, pipeline exists, user has PR permissions

On failure: List all failed checks with remediation steps, abort before any work

---

Phase 1: Intelligent Questions (CRITICAL)

Agents: requirements-analyzer, codebase-scanner, decision-identifier

Generates questions upfront for:

• Technical decisions (auth method, database choice, state management)
• Ambiguity resolution (performance targets, user scope, mobile support)
• Missing requirements (error messages, pagination, rate limits)
• Constraints discovered (existing patterns, compliance, database capacity)
• Risk flags (breaking changes, shared libraries, test gaps)

Outcome: All answers gathered → decisions locked → stored in state.json

---

Phase 2: Preparation

• Branch: Detect existing */${issue_key}-* or create {type}/{issue_key}-{slug}
  • Types: feature (Story/Task), bugfix (Bug), hotfix (Hotfix)
  • Example: feature/PROJ-123-oauth-integration
• Jira: Transition issue to "In Progress"

---

Phase 3: Implementation (6-Phase Protocol)

Uses answers + constraints + technical decisions

EXPLORE: triage-agent, task-enricher, agent-router → context, specialists, dependency graph

PLAN: code-architect → execution plan, task breakdown, risk mitigation + checkpoint (ask user approval)

CODE: Domain specialists + code-reviewer (continuous) → implementation, tests + checkpoint (show diff, ask review)

• Coding standards enforced: terraform (snake_case vars), python (PascalCase classes), TypeScript (camelCase functions), database (snake_case tables)

TEST: test-strategist, test-runner → all tests passing, coverage threshold met

FIX: debugger, fixer (max 3 iterations) → resolve failures, checkpoint if still failing

DOCUMENT: documentation-writer, confluence-manager → README, Confluence, API docs

---

Phase 4: Delivery

Commit: Message format {type}({issue_key}): {summary} with body: decisions made, breaking changes, co-authors

Push & PR: Create PR with summary, technical decisions, auto-generated changes, test results, coverage, Jira link

---

Phase 5: CI Integration

Monitor: build status, tests, coverage, security scan, lint

On failure:

• Build failed: attempt fix (1 retry)
• Tests failed: show failures, prompt /jira:iterate
• Security issues: flag for council review

Collect results: status, duration, coverage delta, security scan output

---

Phase 6: Council Review

Dynamic selection: Auto-include based on files changed

• src/auth/* → security-auditor
• src/api/* → api-reviewer, security-auditor
• *.tsx → accessibility-expert
• prisma/* → database-reviewer
• *.test.* → test-strategist
• Always: code-reviewer

Context: PR diff, CI results, user answers, technical decisions

Finding format: what, why, evidence, impact, fix, verify

Actions: inline comments, decision vote, summary with recommendations

---

Phase 7: Completion

• Update Jira: transition to "In Review", add shipping comment
• Output manifest: issue, branch, PR, duration, agents used, CI status, council decision
• Save state: sessions/ship/${issue_key}/state.json

---

State Machine

States: INITIALIZED → PREFLIGHT_PASSED → QUESTIONS_GATHERED → BRANCH_CREATED → EXPLORE_COMPLETE → PLAN_COMPLETE → CODE_COMPLETE → TESTS_PASSING → PR_CREATED → CI_COMPLETE → COUNCIL_COMPLETE → SHIPPED

Checkpoints: after_plan (ask user), after_code (ask user), auto after PR/council

Resume:

/jira:ship PROJ-123 --status          # Check state
/jira:ship PROJ-123 --resume          # Resume from last checkpoint
/jira:ship PROJ-123 --resume --from=CODE_COMPLETE

---

Commands

/jira:ship PROJ-123 --mode=auto --council=true --wait-ci=true
/jira:ship PROJ-123 --dry-run
/jira:ship PROJ-123 --resume
/jira:ship PROJ-123 --status
/jira:iterate PROJ-123              # Fix council feedback
/jira:council TARGET                # Standalone review

---

Dry Run Output

Shows what WOULD happen: pre-flight checks, questions to ask (~4), branch name, files to modify (~8), estimated lines (300-500), tests to add (~15), PR creation, council members, estimated duration (15-25m), estimated agents (8-10)

---

Error Handling

Error	Recovery
Pre-flight fails	List issues + remediation, abort
User abandons questions	State saved, resume later with --resume
Code phase fails	Checkpoint saved, can resume or rollback
Tests failing	Offer: continue, manual help, abort
CI fails	Offer: /jira:iterate to fix
Council rejects	Offer: /jira:iterate for feedback

---

Configuration

ship:
  version: "2.0"
  preflight: {required: true, timeout: 30s}
  questions: {enabled: true, max_questions: 10}
  checkpoints: {after_plan: ask_user, after_code: ask_user, after_council: auto}
  ci: {wait_for_results: true, timeout: 600s, fail_on_security_critical: true}
  council: {default_preset: standard, dynamic_selection: true, explanation_engine: true, approval_threshold: 0.75}
  state: {persist: true, location: "sessions/ship/", retention_days: 30}

---

Golden Armada | End-to-End Ship Signature