# Stash — Shared Memory for Coding Agents

> Stash is shared memory for coding agents: it captures session transcripts into a workspace and provides a CLI to search and reuse that knowledge.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Install / set up:
   ```bash
   bash -c "$(curl -fsSL https://raw.githubusercontent.com/Fergana-Labs/stash/main/install.sh)"
   ```
2. Start / smoke test:
   ```bash
   stash history search
   ```
3. Verify:
   - Run `stash history search` (or `--help`) and confirm the CLI is installed and can connect to your workspace/memory backend.

## Intro

Stash is shared memory for coding agents: it captures session transcripts into a workspace and provides a CLI to search and reuse that knowledge.

- **Best for:** teams where multiple people/agents work on the same repo and want knowledge to compound across runs
- **Works with:** CLI installs + coding agents (Claude Code/Cursor/Codex/etc.), optional self-host via docker compose
- **Setup time:** 15 minutes

## Practical Notes

- Designed for team-scale memory: multiple agents feed one shared workspace
- Self-host option via `docker compose -f docker-compose.prod.yml up -d` (per README)
- GitHub stars/forks (verified): see Source & Thanks

Stash is useful when “tribal knowledge” becomes your bottleneck:

- One engineer already solved a build issue, but others keep re-solving it.
- Agents generate great notes in transcripts, but the information is lost after the run.
- A long-running agent session becomes faster when it can query prior decisions.

A practical playbook:

1. Start by capturing only the runs you want shared (opt-in).
2. Create one workspace per repo or per product area (keep scope tight).
3. Promote durable facts into the wiki (decisions, commands, common fixes), and let noisy transcripts expire.

This turns agent usage into an asset: later runs start with more context and fewer repeated questions.

### FAQ

**Q: What is Stash?**
A: A shared memory/workspace for coding agent session knowledge, accessible via CLI.

**Q: Can I self-host it?**
A: Yes — the README includes a docker compose production file for self-hosting.

**Q: How do I keep it clean?**
A: Keep scope per repo/team, opt in transcripts, and promote durable facts into the wiki.

## Source & Thanks

> Source: https://github.com/Fergana-Labs/stash
> License: MIT
> GitHub stars: 95 · forks: 30

---

<!-- ZH -->

## 快速使用

1. 安装 / 设置：
   ```bash
   bash -c "$(curl -fsSL https://raw.githubusercontent.com/Fergana-Labs/stash/main/install.sh)"
   ```
2. 启动 / 冒烟测试：
   ```bash
   stash history search
   ```
3. 验证：
   - 运行 `stash history search`（或 `--help`），确认 CLI 已安装并能连接到 workspace/memory 后端。

## 简介

Stash 是面向编码 Agent 的团队记忆层：把多人的会话记录汇入同一 workspace，再通过 CLI 提供检索与复用，让知识在不同成员与不同 Agent 之间持续沉淀与复利增长并可追溯。

- **适合谁：** 多人/多 Agent 协作同一代码库，希望知识在多次 run 里复利增长的团队
- **可搭配：** CLI 安装 + Claude Code/Cursor/Codex 等编码 Agent，可选 docker compose 自建
- **准备时间：** 15 分钟

## 实战建议

- 面向团队记忆：多个 Agent 的 run 汇入同一 workspace
- README 提供自建方式：`docker compose -f docker-compose.prod.yml up -d`
- GitHub stars / forks（已核验）：见「来源与感谢」

当“口口相传的经验”开始拖慢团队时，Stash 的价值会很明显：

- 某个同事解决过一次构建问题，但其他人不断重复踩坑；
- Agent run 的 transcript 里有很多有用信息，但会话结束就散了；
- 长对话 Agent 如果能查询历史决策，效率会显著提升。

一个更务实的使用节奏：

1. 先只收集你愿意共享的 run（opt-in）。
2. 按仓库/产品域划分 workspace（范围越清晰越好）。
3. 把可复用事实沉淀进 wiki（决策、命令、常见修复），把噪声型 transcript 控制住。

这样 Agent 的使用就会从“个人努力”变成“团队资产”，后续 run 的起点更高。

### FAQ

**Stash 是什么？**
答：把编码 Agent 的会话知识沉淀到共享 workspace，并提供 CLI 检索与复用。

**能自建吗？**
答：可以，README 提供了 docker compose 的生产部署文件。

**怎么避免变脏？**
答：控制 workspace 范围、按需上传 transcript，并把耐用事实沉淀进 wiki。

## 来源与感谢

> Source: https://github.com/Fergana-Labs/stash
> License: MIT
> GitHub stars: 95 · forks: 30


---
Source: https://tokrepo.com/en/workflows/stash-shared-memory-for-coding-agents
Author: Agent Toolkit