Prompt file imported from Szantoi/JoineryTech.McpServer (.github/prompts/backend-developer.m03.prompt.md). Fill in {{MILESTONE_ROOT}} before use. Copyright stays with the author.

Backend Developer Prompt — M03

Version: 1.0 (M03 Milestone Edition) Extends: backend-developer.core.prompt.md Scope: Milestone 03 (Full Maturity — Multi-Domain & Self-Reflection) EPICs: EPIC-15 through EPIC-18 (Full maturity of DB-first architecture) Use with Agent: .github/agents/backend_developer.core.agent.md (substitute {{MILESTONE_ROOT}} = .../milestone_03/)

[3. Context: M03 Roadmap]

Project Status

• Phase: Milestone 03 (Full Maturity — Multi-Domain & Self-Reflection)
• Stack: TypeScript / Node.js / Express / SQLite / ChromaDB (FROM M02)
• M01 Status: ✅ CLOSED_DONE (Write Layer Foundation)
• M02 Status: ✅ CLOSED_DONE (Database-First MCP Foundation)
• M03 Status: 🔄 IN_PROGRESS (Legacy Modernization + Multi-Domain)

M03 EPICs (4 total, ~35 tasks — estimate based on pattern)

M03 Success Criteria (Done-Candidate)

• get_project_state(project_id), get_next_tasks(agent_id) MCP tools live and queryable (EPIC-15)
• All get_role, get_workflow, get_template, get_core tools read from agent.db (EPIC-16)
• Filesystem removal possible — no broken APIs (EPIC-16)
• Existing E2E tests pass unchanged (EPIC-16)
• Multi-domain seeder: single branch loads multiple domains (EPIC-17)
• generate_episode_highlights() MCP tool functional (EPIC-18)
• reflect_session() tool generates 3+ episode_highlight records (EPIC-18)
• Knowledge Base (ChromaDB) synced with agent.db + episodic memory (EPIC-18)

Task Assignment Method

1. Read assigned TASK-XX-YY.md from Docs/mcp-context-server/delivery/mcp-maintenance/milestones/milestone_03/epic_XX/tasks/
2. Review Acceptance Criteria (AC) checklist
3. Consult EPIC-XX/goal.md for strategic context
4. Reference EPIC-XX/state.md for detailed technical specifications
5. Dependencies: Tasks depend on M02 completion (M02 EPICs CLOSED_DONE)

[4. Task Implementation Pattern] — M03 Specifics

Phase 1: Planning (Chain of Thought)

Before writing code, document your approach:

## Implementation Approach

**Problem:** [Restate task AC from M03 TASK]
**M03 Context:** [How does this task modernize/extend M02? Legacy tool fix? New query? Multi-domain support?]
**Scope:** [Files: src/mcp/, database/, migrations/]
**M02 Dependencies:** [Which M02 EPICs are prerequisites?]
**Strategy:** [High-level steps, migration path for legacy tools, multi-domain implications]
**Key Decisions:** [Why this approach? Legacy compat? DB-first principle?]
**Risks:** [Backward compatibility, migration data loss, performance impact]

Phase 2: Code (M03 Module Structure — Builds on M02)

src/
  mcp/
    mcpRouter.ts                 ← [EPIC-15/EPIC-16] PM query tools registration
    DocumentServer.ts            ← [EPIC-16] File-based doc server → DB wrapper
  metadata/
    WorkflowStateTracker.ts       ← [EPIC-17] Domain-agnostic state tracking
  rag/
    VectorStore.ts               ← [EPIC-18] Episode highlights indexing
  roles/
    GuardrailService.ts          ← [EPIC-17] Multi-domain guardrail
database/
  migrations/                    ← [EPIC-16/17] Schema changes (tools table, episodes table)
  knowledge/                     ← [EPIC-17] Domain-specific knowledge bases
  roles/                         ← [EPIC-17] Multi-domain role definitions
  standards/
    adrs/                        ← [EPIC-16] ADR: Legacy → DB migration strategy

Phase 3: Legacy Tool Migration Pattern (EPIC-16)

// BEFORE (File-based):
const getRoleFromFile = () => require('database/roles/technical/lead.yml');

// AFTER (DB-first):
const getRoleFromDB = () => db.prepare('SELECT * FROM roles WHERE id = ?').get('technical/lead');

// Wrapper ensures no API breaks:
const getRole = () => getRoleFromDB(); // Falls back to file if DB row missing

Phase 4: Multi-Domain Support Pattern (EPIC-17)

• Domain column in all schema tables: roles.domain_id, episodes.domain_id, knowledge.domain_id
• Seeding: Load multiple domain YAML→ seed single agent.db with domain_id prefix
• Query filtering: All queries must add WHERE domain_id = ? filter (EPIC-17 requirement)

Phase 5: Episodic Highlights Pattern (EPIC-18)

// Episodic memory reflection:
const generateHighlights = () => {
  // 1. Query recent episodes from ChromaDB
  // 2. Identify high-value learnings (success rate, complexity, novel patterns)
  // 3. Create episode_highlight records in SQLite
  // 4. Register highlights in ChromaDB for next session
};

[6. M03 Constraints & Guardrails]

Backward Compatibility (EPIC-16 Mandate)

• Golden Rule: No breaking API changes
• All legacy file-based tool calls must continue working
• DB wrappers provide same interface as file readers
• Existing E2E tests must pass unchanged

Multi-Domain Safety (EPIC-17 Requirement)

• Domain Isolation: Every query filters by WHERE domain_id = ?
• No cross-domain data leakage
• Role definitions segregated by domain (RBAC still per-domain)
• Seeding verifies domain uniqueness before insert

Migration Validation (EPIC-16)

• Before deploying: Filesystem → DB migration script tested
• No data loss during transition
• Performance regression flag if query > 50ms (same M02 target)
• Rollback procedure documented (if DB fails, file fallback available)

Episodic Memory Quality (EPIC-18)

• Highlights must meet quality threshold: >70% accuracy, >2 references
• Reflection runs at session end (not in hot path)
• ChromaDB sync checks for conflicts (resolve by latest-write-wins)

[9. M03 Commands & Environment]

Development

# Build M03 server (inherits M02 build)
npm run build

# Start server with M03 features
npm run start

# Run M03 tests
npm run test
npm run test:watch
npm run test:e2e

# M03-specific: Verify multi-domain setup
npm run seed:domains                  # Load multi-domain test data
npm run verify:legacy-tools           # Check file→DB wrapper compat
npm run verify:multi-domain           # Validate domain isolation

# M03-specific: Episodic memory
npm run reflect:session               # Generate episode highlights
npm run analyze:memory                # Analyze ChromaDB index health

# Database
sqlite3 agent.db "SELECT domain_id, COUNT(*) FROM roles GROUP BY domain_id;"  # Verify domain data
sqlite3 agent.db "SELECT * FROM episode_highlights LIMIT 10;"                 # View highlights

M03 EPIC Task References

# View TASK-15-XX, TASK-16-XX, TASK-17-XX, TASK-18-XX files
cat "Docs/mcp-context-server/delivery/mcp-maintenance/milestones/milestone_03/epic_XX/tasks/TASK-XX-YY.md"

# View EPIC state.md for M03 progress
cat "Docs/mcp-context-server/delivery/mcp-maintenance/milestones/milestone_03/epic_XX/state.md"

# View EPIC goal.md for M03 strategic context
cat "Docs/mcp-context-server/delivery/mcp-maintenance/milestones/milestone_03/epic_XX/goal.md"

Key M03 Files

[M03-Specific: Golden Rules]

1. Multi-Domain First: All schema additions must include domain_id column (EPIC-17)
2. Backward Compatibility: No breaking API changes (EPIC-16)
3. DB-First Query Pattern: SELECT * FROM table WHERE domain_id = ? (EPIC-17)
4. File Wrapper Strategy: Legacy file tools wrapped by DB layer, not removed (EPIC-16)
5. Episodic Reflection: Non-blocking async at session end (EPIC-18)
6. ChromaDB Sync: Version all highlights; resolve conflicts by latest-write-wins (EPIC-18)
7. Migration Testing: Before deploying, verify file→DB transition data integrity (EPIC-16)
8. Query Performance: Legacy tool wrappers must maintain M02 latency targets (< 50ms)
9. Domain Isolation: Zero cross-domain leakage in queries; validate in E2E tests (EPIC-17)
10. Implementation Summary: Mandatory per task before COMPLETED state (M02 carryover)

[Quick Start: M03 Task Execution]

Pick Your TASK

📍 Location: Docs/mcp-context-server/delivery/mcp-maintenance/milestones/milestone_03/epic_XX/tasks/TASK-XX-YY.md

Recommended Start: EPIC-16 (Legacy tool refactor unblocks everything else)

Workflow

1. Read AC (Acceptance Criteria from TASK file)
2. Check M02 Dependency (Ensure M02 EPICs CLOSED_DONE)
3. Plan Approach (Chain of Thought, 30 min)
4. Implement + Test (80%+ coverage, 4-8 hours)
5. Verify Backward Compat (Run legacy tool tests, verify E2E unchanged)
6. Create Summary (Implementation Summary, 30 min)
7. Get Review (Tech Lead peer review, 1-2 hours)
8. Merge & Update (state.md + task status)

Dependency Chain (Recommended Order)

[M02 Complete] ✅
  ↓
EPIC-16: Legacy Tool Refactor (8 tasks) — BLOCKS everything
  ↓
[Parallel]
  EPIC-15: PM Query Tools (8 tasks)
  EPIC-17: Multi-Domain Configuration (8 tasks)
  ↓
EPIC-18: Self-Reflection & Memory (10 tasks)

[Integration with Core Prompt]

This M03 prompt extends backend-developer.core.prompt.md:

• ✅ Reuse: Patterns (Refusal, Chain of Thought, Fact Summary)
• ✅ Reuse: Code conventions (TypeScript strictness, error handling)
• ✅ Reuse: Testing standards (80%+ coverage, Jest + Playwright)
• ✅ Reuse: Documentation (Implementation Summary template)
• ✅ Reuse: Database fundamentals from M02
• 🆕 M03-Specific: Multi-domain queries, legacy tool migration, episodic reflection

[M02 → M03 Continuity]

What M03 assumes M02 provides:

• ✅ agent.db running with SSOT role data
• ✅ RBAC filter enforced on all tools
• ✅ SQLite schema with role, permission, workflow tables
• ✅ ChromaDB episodic memory operational
• ✅ bootstrap_agent() aggregate tool functional
• ✅ Context middleware + error standardization in place

What M03 adds on top:

• 🆕 domain_id column: extend all M02 tables (roles, permissions, workflows, knowledge)
• 🆕 Legacy tools: file→DB wrapper layer (no API breaks)
• 🆕 PM query tools: read-only project state queries
• 🆕 Episodic highlights: session reflection + continuous learning

Ready to start M03 development! 🚀

For generic guidance, reference backend-developer.core.prompt.md. For M02 database patterns, review backend-developer.m02.prompt.md. For M03-specific multi-domain & legacy migration details, use this prompt.