
kasfulk/chain-of-truth
Productivity workflowこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
Validated artifacts as the Source of Truth (SoT) — not prompts.
概要
Validated artifacts as the Source of Truth (SoT) — not prompts. A workflow for AI-assisted software development that replaces prompt-driven development with Source-of-Truth-driven development. Instead of scattering requirements across chat conversations, CoT produces validated artifacts that carry project knowledge, making prompts simple execution instructions. SRS → Product Structure (IA + Design System + User Flows) → HiFi Prototype → Data Model → UCIC → Implementation → Testing AI optimizes , not prompts. When context (validated artifacts) is reliable, prompts become simple. When context is unreliable, prompt engineering is endless. - Building a new feature or application with AI - Planning a project spanning multiple sessions - Reducing inconsistency in AI outputs - Structuring requirements traceably 1. Run the skill: /chain-of-truth 2. Follow the 7 phases — each phase has dependencies (e.g. Data Model comes from User Flows, not from the SRS) 3.
README
以下は原文 README の抜粋です。原文を見る。
Chain of Truth
Validated artifacts as the Source of Truth (SoT) — not prompts.
A workflow for AI-assisted software development that replaces prompt-driven development with Source-of-Truth-driven development. Instead of scattering requirements across chat conversations, CoT produces validated artifacts that carry project knowledge, making prompts simple execution instructions.
7 phases: SRS → Product Structure (IA + Design System + User Flows) → HiFi Prototype → Data Model → UCIC → Implementation → Testing
Philosophy
AI optimizes context, not prompts. When context (validated artifacts) is reliable, prompts become simple. When context is unreliable, prompt engineering is endless.
Traditional vibe coding:
Requirement → Prompt → AI → Prompt Revision → AI → More Prompts → AI
(project knowledge scattered across conversations)
Chain of Truth:
Requirement → Source of Truth → AI
(SoT carries knowledge, prompts become execution instructions)
7 Phases & Source of Truth Chain
SRS (SoT #1)
↓
IA (SoT #2) + Design System (SoT #3) + User Flows (SoT #4) [parallel]
↓
HiFi Prototype (SoT #5)
↓
Data Model (SoT #6) [from User Flows, NOT directly from SRS]
↓
UCIC (SoT #7) [from User Flows + Data Model]
↓
Implementation [Frontend: HiFi+UCIC, Backend: DataModel+UCIC]
↓
Testing [from User Flows + UCIC, not from code]
When to Use
- Building a new feature or application with AI
- Planning a project spanning multiple sessions
- Reducing inconsistency in AI outputs
- Structuring requirements traceably
Quick Start
- Run the skill:
/chain-of-truth - Follow the 7 phases in order — each phase has dependencies (e.g. Data Model comes from User Flows, not from the SRS)
- Use the templates in
templates/to generate artifacts - Validate each artifact before treating it as a Source of Truth for downstream work
- Implement with minimal prompts that reference artifacts, not describe features
Expected docs/ layout
docs/
srs.md ← SoT #1
information_architecture.md ← SoT #2
design_system.md ← SoT #3
user_flows/
uc-001-[name].md ← SoT #4 (one file per use case)
data_model.md ← SoT #6
system_logics/ ← SoT #7 (UCIC, one per use case)
uc-001-[name].md
test_plan.md
test_cases.md
test_execution_sheet.md
Structure
SKILL.md — Full skill definition, phase details, traceability, evaluation metrics
templates/ — Templates for each artifact type (SRS, IA, design system, user flow,
data model, UCIC, test plan, test cases, test execution sheet)
.headroom/ — Persistent memory store
README.md — This file
See SKILL.md for complete phase details, traceability paths, revision loops,
pitfalls, and evaluation metrics.
Credit
This skill is based on the Chain of Truth academic research by:
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add kasfulk/chain-of-truth