Harness
개요: Origemite Archive용 휴대용 AI 작업 뼈대입니다. 잡동사니 메모장이 아니며, 에이전트가 같은 Markdown 지식과 템플릿을 골라 쓰게 합니다. 통과 가능한 코딩의 기준을 구현 전에 제시합니다.
workflows/acceptance.md
Portable AI engineering harness for Origemite Archive.
Helps coding agents (Cursor, Codex, Claude Code, …) reuse the same Markdown knowledge and templates in real projects.
Layout
| Path | Question it answers |
|---|---|
orchestration.md | Meta: select / gap / author recipes |
orchestrations/ | Situation recipes (agents + knowledge assembly) |
entrypoints/ | Product adapters (Cursor / Codex / …) |
agents/ | Role contracts (Master writes orchestrations; others execute) |
tools/ | Verification tools the Tester may use to prove vibe-coded results (tests, Playwright, agent-driven browser) |
context/ | Identity, principles, conventions, precedence |
workflows/implement.md | 기본 개발 흐름. 매번 모든 역할을 실행하지 않습니다 |
CURRENT.md | 현재 상태. 작업 로그 전체가 아닙니다 |
history/ | 작업 기록. 필요할 때만 해당 월을 읽습니다 |
profiles/ | Usual asset combo for a kind of stack |
bundles/ | Optional read-list helper (subset of a task orchestration) |
templates/ | Implementation reference assets (code + README) |
Related concepts
- Index (
index/) — what exists where - Templates (
docs/harness/templates/) — concrete implementation shapes - Docs (
docs/) — knowledge, architecture, patterns, ADRs, …
Harness structure must not depend on a single AI vendor. Entrypoints are thin adapters.
Progressive loading
Index (what exists)
→ Master selects orchestration for situation
or gap feedback → write orchestration
→ workers load only that recipe’s knowledge assembly
History / Postmortems only if the selected orchestration lists them.