Vendor-neutral, AI 코딩 에이전트에 대한 지속적인 프로젝트 컨텍스트. 공유 상태는 삶의 docs/context/, Git에서 검토 할 수, 그리고 채팅 리셋, 컴팩트, 및 스위칭 에이전트 (Claude Code, Codex, Antigravity)를 생존.

배치

docs/context/:

기술이 교육 구동되지 않습니다 - 설치하거나 실행 할 수있는 프로그램이 없습니다. 에이전트는 자체 파일 도구와 함께 모든 작업을 수행 date/git]/find] (선택적으로 [T5]]), SKILL.md 및 참조/procedures.md 뒤에. 비밀, 원시 성적을 저장하지 않거나, Chain-of-thought.

설치

와 npx skills (추천 — 모든 에이전트에 대한 작품)

대안 — symlink

어느 방법, 스킬의 Initialize 동작을 한 번 대상 repo - 그것은 비계 docs/context/, AGENTS.md]/CLAUDE.md]/GEMINI.md 포인터, /ac-] 슬래시 명령, .gitignore] 항목. 네이티브 설치는 선택 사항입니다. 기술 또한 "제너콘텍스 기술 사용"에 어떤 에이전트를 말하는 것에 의해 작동 - 그것은 명시적 인 주장을 위해 설계되었습니다, 설명 매칭하지.

agent-context-kit

Vendor-neutral, persistent project context for AI coding agents. Shared state lives under docs/context/, stays reviewable in Git, and survives chat resets, compaction, and switching agents (Claude Code, Codex, Antigravity).

Layout

skills/engineering/agent-context/   # the installable skill (SKILL.md + references/ + assets/)
docs/context/                       # per-project context this skill manages
AGENTS.md / CLAUDE.md / GEMINI.md   # short pointers so any agent auto-discovers docs/context/

docs/context/:

docs/context/
├── index.md                # generated routing table
├── summary.md              # generated rolling summary — read this first
├── project.md              # durable project-wide context
├── backlog.md              # things to do next, captured with /ac-note
├── config.yaml             # tracking and summary policy
├── topics/                 # reusable, evidenced knowledge
├── work/                   # active task state and checkpoints
├── archive/                # completed task records
└── generated/repo-tree.md  # reproducible repository structure

.context.local/             # gitignored machine-local/private context

The skill is instruction-driven — no program to install or run. The agent performs every action with its own file tools plus date/git/find (optionally tree), following SKILL.md and references/procedures.md. Never stores secrets, raw transcripts, or chain-of-thought.

Install

With npx skills (recommended — works for every agent)

# Claude Code
npx skills add ThanakritAof/agent-context-kit -a claude-code

# Other agents (codex, antigravity, gemini-cli, cursor, etc.)
npx skills add ThanakritAof/agent-context-kit -a 

Alternative — symlink

ln -s "$(pwd)/skills/engineering/agent-context" .claude/skills/agent-context   # Claude Code
ln -s "$(pwd)/skills/engineering/agent-context" .agents/skills/agent-context   # Codex

Either way, tell the agent to run the skill’s Initialize action once per target repo — it scaffolds docs/context/, the AGENTS.md/CLAUDE.md/GEMINI.md pointers, the /ac- slash commands, and the .gitignore entry. Native install is optional: the skill also works by just telling any agent to “use the agent-context skill” — it’s designed for explicit invocation, not description-matching.

If you installed the skill globally (as npx skills add does by default), also drop /ac-init at the same personal scope so it works in any repo without a prior Initialize run there:

mkdir -p ~/.claude/commands
curl -fsSL https://raw.githubusercontent.com/ThanakritAof/agent-context-kit/main/skills/engineering/agent-context/assets/templates/commands/ac-init.md \
  -o ~/.claude/commands/ac-init.md

Continuity hook (optional)