HALO Workflow v3

⚙

Workflow Process — 9 Phases + JUDGE

P1~P7, P9 Main Direct P8 Review ×3 JUDGE ×1 LOOPBACK Recovery

Main

📋

Requirements
+ Constraint

Main

🔍

Exploration
Direct

Main

📐

Architecture
Direct

Main

🔴

Unit Test
TDD RED

Main

🟢

Implement
TDD GREEN

Main

🔀

IT & E2E
Real Env

Main

▶

Test Run
+ Smoke

×3

🔎

Code
Review

×1

🔄

JUDGE

RTM Only
Loopback

Main

📑

Report

P1 init

P4 +UT

P5 +impl

P6 +IT/E2E

P7 result

P8 review

JUDGE

🤖

Sub-Agents — Review (P8) + JUDGE only

P8: Review ×3 JUDGE: RTM ×1 File paths only, no summary

P8 Code Review — 3 Perspectives

Agent 1: Quality

DRY, readability,
maintainability

Agent 2: Bugs

Logic errors,
edge cases

Agent 3: Security

OWASP, conventions,
vulnerabilities

Each reports issues (80%+ confidence) → Main reflects to RTM Review column

JUDGE RTM-Based LOOPBACK Evaluation

Input: RTM only

REQ ↔ TC mapping
Impl location (file:line)
Result (PASS/FAIL)
Review (MAJOR/CRITICAL)

Process

1. Find FAIL/CRITICAL in RTM
2. Trace TC → REQ → file:line
3. Read code if needed
4. Classify root cause

Output

PASS → P9
LOOPBACK → P3/P4/P5/P6
+ root cause + instructions

Reads RTM → classifies root cause → Main executes LOOPBACK or proceeds to P9

📂

Artifacts — Product (permanent) + Checkpoint (temporary)

RTM = Single Source of Truth

permanent

📄 docs/

requirements/feature.md
requirements/feature-rtm.md
architecture/feature.md

★ RTM = Source of Truth

permanent

🔧 tests/

unit/feature.test.*
integration/feature.int.*
e2e/feature.e2e.* Real Env

permanent

💻 src/

feature/*.*
implementation code

temp / gitignored

⚙ .workflow/

state.json
loopback-context.md
phase-results/P*.md

permanent

📑 reports/

feature-completion.md
test + review summary