Esta página ainda está em inglês. A tradução para português está em andamento.

Engineering Layer

Attribution: the Engineering Layer is derived from oh-my-claudecode (MIT, by Yeachan Heo). EvoNexus imports the agents and adapts them to the workspace conventions (themed names, EvoNexus standard pattern, workspace/development/ paths). See NOTICE.md at the repo root for the full list of derived components and modifications.

Native additions (v0.13.0): helm-conductor and mirror-retro are native EvoNexus agents, not derived from oh-my-claudecode. Helm orchestrates the engineering cycle and routes tasks to phase owners; Mirror runs blameless retrospectives at the end of features and sprints.

What It Is

The Engineering Layer is a complete software development team built into EvoNexus. 21 specialized agents (19 derived + 2 native) + 25 dev-* skills + 15 templates + a canonical 6-phase workflow (dev-phases.md) that cover the full dev lifecycle: discovery, planning, solutioning, build, verify, retro. It exists as a layer ortogonal to the Business Layer (Clawdia, Flux, Atlas, etc.). Business agents handle operations (emails, finance, community); engineering agents handle code (architecture, bugs, reviews, tests). Cross-layer handoffs are common — e.g., Nova writes a PRD → Apex reviews the architecture → Bolt implements → Lens reviews → Oath verifies → Mirror captures lessons.

When to Use

You write code on top of EvoNexus
You want a structured dev workflow with separation of concerns (planner ≠ executor ≠ reviewer ≠ verifier)
You want code review, security audits, and verification as first-class agents
You want autonomous end-to-end execution from a brief to working code (dev-autopilot)

If you only need EvoNexus for ops/business work, you can ignore the Engineering Layer entirely — it doesn’t get in the way.

The 21 Agents

Reasoning (opus / sonnet) — 8 agents

These agents do deep cognitive work: design, analysis, planning, critique, reflection. Higher token cost, higher quality output.

Agent	Slash	Role	READ-ONLY?
Apex	`/apex-architect`	Architect — architectural design, debugging diagnosis, tradeoffs with file:line citations	✅
Echo	`/echo-analyst`	Analyst — discovery, gap analysis, hidden assumptions, missing acceptance criteria	✅
Compass	`/compass-planner`	Planner — interview-driven 3-6 step work plans + PRD production with explicit ADRs
Raven	`/raven-critic`	Critic — multi-perspective adversarial review with pre-commitment predictions	✅
Lens	`/lens-reviewer`	Code Reviewer — 2-stage review (spec compliance + quality), OWASP, SOLID, severity ratings	✅
Zen	`/zen-simplifier`	Code Simplifier — deslop, reduce nesting, eliminate redundancy without changing behavior
Vault	`/vault-security`	Security Reviewer — OWASP Top 10, secrets scan, dependency audit, prioritized remediation	✅
Mirror ⭐	`/mirror-retro`	Retrospective — blameless lessons learned, reads the full feature folder, proposes memory updates

Execution (sonnet) — 11 agents

These agents do focused implementation work: writing code, running tests, investigating bugs, producing artifacts, sequencing work.

Agent	Slash	Role
Bolt	`/bolt-executor`	Executor — precise multi-file implementation with smallest viable diff
Hawk	`/hawk-debugger`	Debugger — root cause analysis, build error fixing, 3-failure circuit breaker
Grid	`/grid-tester`	Test Engineer — TDD discipline, test pyramid (70/20/10), flaky test diagnosis
Probe	`/probe-qa`	QA Tester — interactive tmux sessions, real commands, real output, always cleans up
Oath	`/oath-verifier`	Verifier — evidence-based completion checks, never trusts “should work”
Trail	`/trail-tracer`	Tracer — causal investigation with competing hypotheses, ranks by evidence strength
Flow	`/flow-git`	Git Master — atomic commits, style detection, safe rebasing
Scroll	`/scroll-docs`	Document Specialist — external SDK/API docs lookup with citations
Canvas	`/canvas-designer`	Designer — production-grade UI/UX, framework-idiomatic, distinctive aesthetic
Prism	`/prism-scientist`	Scientist — formal statistical analysis with CI, effect size, p-value
Helm ⭐	`/helm-conductor`	Conductor — cycle orchestration, sequencing, routing to phase owners, sprint planning

⭐ = native EvoNexus agent (not derived from oh-my-claudecode)

Speed (haiku) — 2 agents

Fast, parallel, low-token agents for high-volume work.

Agent	Slash	Role
Scout	`/scout-explorer`	Explorer — parallel codebase search (Glob/Grep), absolute paths, file:line evidence
Quill	`/quill-writer`	Writer — quick technical docs, README, comments, every code example tested

The 6-Phase Workflow

The engineering layer follows a canonical workflow documented in .claude/rules/dev-phases.md. It is a guide, not a rigid gate — simple changes can skip phases, complex features should follow them in order.

Discovery → Planning → Solutioning → Build → Verify → Retro

Phase	Owner	Output
1. Discovery	`@echo-analyst`	`[C]discovery-{feature}.md` — gaps, assumptions, open questions
2. Planning	`@compass-planner`	`[C]prd-{feature}.md` + `[C]plan-{feature}.md`
3. Solutioning	`@apex-architect`	`[C]architecture-{feature}.md` (ADR format)
4. Build	`@bolt-executor`	code + tests + commits
5. Verify	`@oath-verifier`	`[C]verification-{feature}.md` — evidence-based PASS/FAIL
6. Retro	`@mirror-retro`	`[C]retro-{feature}.md` — lessons + memory updates

Cycle orchestration — @helm-conductor sits above the phases and answers “what next?”, “who does it?”, and “what’s blocked?”. Call Helm when you have multiple active features or need sprint sequencing. Feature folders — non-trivial work lives in workspace/development/features/{feature-slug}/ where all artifacts of one feature are grouped together. Standalone artifacts continue to live in workspace/development/{type}/.

The 25 `dev-*` Skills

Skills that orchestrate the engineering agents.

Tier 1 — Core orchestration (15)

dev-autopilot — full lifecycle: spec → plan → code → QA → validation
dev-plan — interview-driven planning (delegates to compass-planner)
dev-ralplan — multi-perspective consensus (Planner + Architect + Critic)
dev-deep-interview — Socratic Q&A to crystallize requirements
dev-deep-dive — causal trace + requirements crystallization combined
dev-external-context — parallel external doc lookups
dev-trace — evidence-driven causal investigation
dev-verify — evidence-based completion verification
dev-ultraqa — QA cycling loop (build/lint/test/fix up to 5 cycles)
dev-visual-verdict — visual regression testing
dev-ai-slop-cleaner — post-AI cleanup (single-use helpers, over-abstractions)
dev-sciomc — scientific method scaffolding
dev-team — on-demand parallel agent spawning
dev-ccg — tri-model orchestration (Claude + Codex + Gemini)
dev-ralph — persistence loop with iteration-based circuit breaker

Tier 2 — Setup & infrastructure (5)

dev-mcp-setup — configure MCP servers
dev-deepinit — generate hierarchical AGENTS.md / CLAUDE.md for new projects
dev-project-session-manager — git worktrees per issue/PR
dev-configure-notifications — Telegram/Discord/Slack alerts for eng events
dev-release — generic release prep (changelog, version bump, tag)

Tier 3 — Meta / utilities (5)

dev-cancel — clean stop of any active eng workflow
dev-remember — quick context persistence (lighter than mempalace)
dev-ask — single-shot query to a specific model (Claude/Codex/Gemini)
dev-learner — extract reusable skills from conversation patterns
dev-skillify — convert current conversation into a new skill

Recommended Pipelines

End-to-end implementation

User: "build me a CRUD API for a bookstore inventory in TypeScript"
   ↓
dev-autopilot
   ├── Phase 0: echo-analyst → spec
   ├── Phase 1: compass-planner → plan
   ├── Phase 2: bolt-executor → code (parallel where safe)
   ├── Phase 3: dev-ultraqa → build/test/fix loop
   ├── Phase 4: lens-reviewer + vault-security + apex-architect → validation
   └── Phase 5: oath-verifier → final evidence-based verdict

High-stakes plan

User: "we need to migrate auth from JWT to session cookies"
   ↓
dev-ralplan (consensus mode)
   ├── compass-planner → initial plan with principles, drivers, options
   ├── apex-architect → consensus addendum (steelman antithesis, tradeoff tension)
   └── raven-critic → adversarial review (gap analysis, severity ratings)
   ↓
Final plan with ADR + pre-mortem (deliberate mode)

Bug investigation

User: "the bot drops messages but only on Tuesdays"
   ↓
trail-tracer (multi-hypothesis)
   ├── Generate 2+ competing explanations
   ├── Collect evidence FOR and AGAINST each
   ├── Apply lenses (systems / premortem / science)
   └── Identify the discriminating probe
   ↓
hawk-debugger (once root cause is clear)
   ├── Reproduce
   ├── Minimal fix
   └── 3-failure circuit breaker
   ↓
oath-verifier (regression check)

Pre-merge gate

git diff
   ↓
lens-reviewer (code quality, severity-rated)
   ↓
vault-security (OWASP audit)
   ↓
dev-ultraqa (build/test/fix until green)
   ↓
oath-verifier (acceptance criteria mapping)
   ↓
flow-git (atomic commits matching project style)

Working Folder

All engineering layer artifacts persist in workspace/development/:

workspace/development/
├── architecture/      — apex-architect: ADRs, design tradeoffs, debug analyses
├── plans/             — compass-planner: work plans, RALPLAN-DR consensus, open questions
├── specs/             — echo-analyst, dev-deep-interview: gap analyses, requirement specs
├── reviews/           — lens-reviewer, vault-security: code reviews, security audits
├── debug/             — hawk-debugger, trail-tracer: bug reports, trace investigations
├── verifications/     — oath-verifier, grid-tester, probe-qa: verification reports, test strategies
├── research/          — scroll-docs, prism-scientist, scout-explorer: external research, statistical analyses
└── README.md

Distinction from workspace/projects/:

workspace/development/ — engineering layer artifacts (analyses, plans, reports). Owned by Claude.
workspace/projects/ — active git projects (Evolution API, Evo AI, Evo Go). Owned by you. Engineering agents read for context; only bolt-executor writes code there, and only with your approval.

Templates

Each engineering agent has a template at .claude/templates/dev-*.md for its primary output:

Agent	Template
apex-architect	`dev-architecture-decision.md`
compass-planner	`dev-work-plan.md`
lens-reviewer	`dev-code-review.md`
hawk-debugger	`dev-bug-report.md`
oath-verifier	`dev-verification-report.md`
echo-analyst, dev-deep-interview	`dev-deep-interview-spec.md`
vault-security	`dev-security-audit.md`
grid-tester	`dev-test-strategy.md`
trail-tracer	`dev-trace-report.md`
scout-explorer	`dev-explore-report.md`
canvas-designer	`dev-design-spec.md`
prism-scientist	`dev-analysis-report.md`
scroll-docs	`dev-research-brief.md`
raven-critic	`dev-critique.md`
zen-simplifier	`dev-simplification-report.md`

Memory

Each engineering agent has its own memory folder at .claude/agent-memory/{agent-name}/. Agents update memory with patterns worth carrying forward (anti-patterns, gotchas, codebase conventions). Engineering agents use a lighter memory pattern than business agents — no rich auto-memory block with user/feedback/project/reference types. They are tools, not “colleagues” — Apex doesn’t need to learn “the Davidson way”, it just needs to remember the architecture patterns of this codebase.

Cross-Layer Handoffs

The two layers cooperate naturally:

Trigger	Business → Engineering
`@nova` writes a PRD	→ `@apex-architect` reviews the architecture
`@dex-data` finds a metric anomaly	→ `@trail-tracer` investigates causally
`@atlas-project` flags an open PR	→ `@lens-reviewer` reviews it
`@lex-legal` flags a compliance gap in code	→ `@vault-security` audits
`@mako-marketing` needs a landing page	→ `@canvas-designer` builds it

Attribution

The Engineering Layer is derived from oh-my-claudecode:

Author: Yeachan Heo
Version imported: v4.11.4
Date imported: 2026-04-10

EvoNexus modifications:

Renamed agents to themed names (architect → apex-architect)
Added dev- prefix to all skills for namespace consistency
Adapted memory structure to match EvoNexus per-agent pattern
Removed runtime dependencies on OMC TypeScript src/ (kept only the markdown definitions)
Discarded skills that overlapped with EvoNexus builtins or were OMC meta-skills

See NOTICE.md for the complete attribution and modification list.

Agentes

Skills

Rotinas

Heartbeats e Goals

Dashboard

Guias

Integrações

Provedores

Casos Reais

Referência

Engineering layer

Engineering Layer

What It Is

When to Use

The 21 Agents

Reasoning (opus / sonnet) — 8 agents

Execution (sonnet) — 11 agents

Speed (haiku) — 2 agents

The 6-Phase Workflow

The 25 `dev-*` Skills

Tier 1 — Core orchestration (15)

Tier 2 — Setup & infrastructure (5)

Tier 3 — Meta / utilities (5)

Recommended Pipelines

End-to-end implementation

High-stakes plan

Bug investigation

Pre-merge gate

Working Folder

Templates

Memory

Cross-Layer Handoffs

Attribution

​Engineering Layer

​What It Is

​When to Use

​The 21 Agents

​Reasoning (opus / sonnet) — 8 agents

​Execution (sonnet) — 11 agents

​Speed (haiku) — 2 agents

​The 6-Phase Workflow

​The 25 dev-* Skills

​Tier 1 — Core orchestration (15)

​Tier 2 — Setup & infrastructure (5)

​Tier 3 — Meta / utilities (5)

​Recommended Pipelines

​End-to-end implementation

​High-stakes plan

​Bug investigation

​Pre-merge gate

​Working Folder

​Templates

​Memory

​Cross-Layer Handoffs

​Attribution

Engineering Layer

What It Is

When to Use

The 21 Agents

Reasoning (opus / sonnet) — 8 agents

Execution (sonnet) — 11 agents

Speed (haiku) — 2 agents

The 6-Phase Workflow

The 25 `dev-*` Skills

Tier 1 — Core orchestration (15)

Tier 2 — Setup & infrastructure (5)

Tier 3 — Meta / utilities (5)

Recommended Pipelines

End-to-end implementation

High-stakes plan

Bug investigation

Pre-merge gate

Working Folder

Templates

Memory

Cross-Layer Handoffs

Attribution