Engineering Workflow

engineering-workflow

A 3-tier micro-agent orchestration plugin for architecture decisions across four systems: DB, BE, IF, and SE.

Instead of a monolithic response, it classifies a query, routes it to the right system orchestrators and micro agents, resolves constraints, and returns a unified recommendation.

Quick Start

# Example 1) General query (default: deep)
/engineering-workflow "Which is better for write-heavy workloads: B-tree or LSM-tree?"

# Example 2) Analyze mode
/engineering-workflow "analyze: bottlenecks in our current sharding strategy"

# Example 3) Compare mode
/engineering-workflow "compare: PostgreSQL vs CockroachDB for multi-tenant"

# Example 4) Recommend mode + depth
/engineering-workflow "recommend: caching strategy for read-heavy APIs" --depth shallow

Argument hint:

[query | analyze | compare | recommend] [--domain db|be|if|se] [--depth shallow|deep]

---

Core Features

• Deterministic fast-path classification via scripts/classify-query.sh
• 3-tier orchestration
  • Tier 1: Gateway Router (SKILL)
  • Tier 2: System Orchestrators (DB/BE/IF/SE)
  • Tier 3: Domain Micro Agents
• Constraint propagation and conflict resolution via resolve-constraints.sh
• Quality gates via audit-analysis.sh (confidence/schema/synthesis checks)
• Session learning cache (pattern-cache.json) for repeated query patterns
• False-positive resistant keyword matching using word-boundary based detection
• Exit Checklist quality gates across all agents — checkbox-based output validation ensuring JSON schema conformance, field presence/cardinality, confidence bounds (0.0–1.0), and graceful degradation on insufficient input

---

Routing Overview

System Routing

• DB: storage, index, query plans, concurrency, schema, replication, etc.
• BE: API, service layer, resilience, test architecture, etc.
• IF: infrastructure, deployment, k8s, CI/CD, observability, etc.
• SE: authentication, encryption, RBAC/ABAC, compliance, security, etc.

Sub-routing

• DB Domains A–F: Storage, Query Plan, Concurrency, Schema, I/O, Distributed
• BE Clusters S/B/R/T: Structure, Boundary, Resilience, Test

For detailed matrices and routing rules, see:

• skills/engineering-workflow/SKILL.md
• skills/engineering-workflow/resources/routing-protocol.md
• skills/engineering-workflow/resources/orchestration-protocol.md

---

Execution Pipeline

1. Phase 0 — Classification
2. Phase 1 — Orchestrator Dispatch
3. Phase 2 — Agent Execution
4. Phase 2.5 — Analysis Quality Gate
5. Phase 3 — Constraint Resolution
6. Phase 3.5 — Contract Enforcement Gate
7. Phase 4 — Synthesis (cross-system only)
8. Phase 4.5 — Synthesis Validation Gate
9. Phase 5 — Output Formatting

---

Scripts

Location: plugins/engineering-workflow/skills/engineering-workflow/scripts/

Script	Purpose
classify-query.sh	Query classification (systems, domains, be_clusters, confidence)
resolve-constraints.sh	Constraint conflict detection and resolution
format-output.sh	Final output formatting
validate-agent-output.sh	Agent output schema/quality validation
audit-analysis.sh	Audit tier, confidence, orchestrator, synthesis validation
_common.sh	Shared utilities (keywords/session cache/cleanup)

---

Tests

Location: plugins/engineering-workflow/skills/engineering-workflow/tests/

cd plugins/engineering-workflow/skills/engineering-workflow

# Classification regression tests
bash tests/run-classification-tests.sh

# Constraint resolution tests
bash tests/run-constraint-tests.sh

# Output format tests
bash tests/run-format-tests.sh

# Agent output validation tests
bash tests/run-validation-tests.sh

---

Hook Behavior (.claude-plugin/plugin.json)

• PreToolUse (Edit/Write): blocks edits to production secret/credential-like files
• PostToolUse (Write): validates constraints.json as valid JSON
• Stop: archives constraints and trims history retention

---

Cache and Session Storage

~/.claude/cache/engineering-workflow/

• constraints.json: current session constraint set
• session-history.jsonl: query classification history
• pattern-cache.json: repeated pattern cache
• history/: archived session constraints

---

Directory Structure

plugins/engineering-workflow/
├── .claude-plugin/
│   └── plugin.json
└── skills/engineering-workflow/
    ├── SKILL.md
    ├── agents/
    ├── resources/
    ├── references/
    ├── scripts/
    ├── tests/
    └── templates/

---

Requirements

• bash 3.2+
• jq
• grep, awk, sed
• Unix-like environment (macOS/Linux recommended)

No external build runtime is required; the plugin is script-driven.

---

References

• Main skill definition: plugins/engineering-workflow/skills/engineering-workflow/SKILL.md
• Priority rules: plugins/engineering-workflow/skills/engineering-workflow/resources/priority-matrix.md
• Synthesis protocol: plugins/engineering-workflow/skills/engineering-workflow/resources/synthesis-protocol.md
• Error handling: plugins/engineering-workflow/skills/engineering-workflow/resources/error-playbook.md