Stash plugin
Open infrastructure for agents that learn from experience: session traces, a knowledge base and skill packs via CLI, MCP and API
- 340 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- With recording on, the agent's work history goes to a Stash server
- Sessions can contain code, secrets and personal data that end up in history and the wiki
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/fergana-labs-stashDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- With recording on, the agent's work history goes to a Stash server
- Sessions can contain code, secrets and personal data that end up in history and the wiki
This is third-party code. Review the repository files before installing.
What it does
Stash collects the work history of agents and turns it into knowledge that later runs can see. Hooks for coding agents record prompts, tool calls and responses when session recording is on. Events from your own agents can be sent through the SDK or the API. A scheduled curator reads new sessions and updates linked pages in the Memory wiki, so the takeaway stays available after the original session ends. Agents search and read this knowledge through the CLI, MCP, the API or a virtual file system. Related instructions and files are packaged into a skill: a folder with a `SKILL.md` file. The `stash skills install` command installs skills into an agent, and they update at session start. The repository also holds a research part: trace review, training a reward model from reviewer comments and user reactions, and skill optimization with GEPA. For Claude Code there is a plugin that connects the agent to Stash workspaces, streams activity to history and injects agent memory.
Who it is for. For teams and solo developers who run Claude Code or Codex regularly and want to build up project knowledge in a shared wiki.
Good fit when
- A reviewer corrected the agent and you want to turn the fix into a rule for future runs
- The work history of Claude Code or Codex should be saved and found by search
- Accumulated project knowledge should stay in a wiki after the session ends
- A set of instructions should be packaged as a skill and shared with several agents
- You want to deploy such infrastructure yourself
Not a fit when
- You need only light local memory in one file: Stash deploys a backend, background workers and a database
- Sessions contain data that must not be recorded and sent to a Stash server
- You need guarantees of better quality: the author advises checking the effect on later tasks
Example request
Find the database migration rules in Stash memory and prepare a checklist for the next schema changeLimitations
The reward-model experiment is enabled for accounts created after the rollout. Existing accounts keep their current experience, and access is checked in both the UI and the API. Training needs at least two grounded comparisons, and the AI assessor's judgment reflects model preferences and does not measure customer satisfaction.
The 49% speedup for long Claude Code runs comes from the authors' internal experiment with its own setup, so results on your tasks may differ. Self-hosting means running the backend, background workers and a database yourself.
How to disable. Disable or remove the plugin through `/plugin` in Claude Code and switch off session recording in the Stash settings. Data on the server is deleted with the tools of your workspace.
FAQ
How do I install Stash in Claude Code?
Add the marketplace with `/plugin marketplace add fergana-labs/stash` and install the plugin with `/plugin install stash@stash-plugins`. The plugin connects Claude Code to Stash workspaces and streams activity to history.
Do I need to switch on session recording separately?
Yes, recording is a setting. While it is off, the hooks capture nothing, and events from your own agents can still be sent through the SDK or the API.
What is a skill in Stash?
A folder in the Files section with a `SKILL.md` file that has name and description in its frontmatter. Extra md files belong to the skill. stash_list_skills and stash_read_skill serve skills over MCP.
How do skills reach an agent?
With the `stash skills install` command. Installed skills update at session start, so changes to shared instructions reach the next run.
Can feedback be turned into a skill automatically?
Yes, on a trained reward model choose View skill: GEPA runs on the selected traces, and you download the result as `SKILL.md`.
Discussion
No questions or reviews yet.
Related
PluginsShared cloud memory for a team's agents: stores session traces and turns repeated patterns into skills
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
Stash gives an agent long-term memory in PostgreSQL: entries are merged into facts, relations and patterns, served over MCP, started in Docker
A portable .agent folder with memory, skills and protocols that plugs into many coding agents and keeps its knowledge when you switch tools
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
A tool that builds tested skills from a real workflow and governs their lifecycle through your own marketplace
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Stash (this entry) | no | 340 | ||
| Hivemind | no | 1,641 | shared cloud memory for a team: stores session traces and turns them into knowledge | |
| Basic Memory | no | not required | 4,403 | long-term memory in local Markdown files with links between notes |
| MCP Memory Service | no | not required | 2,054 | a self-hosted long-term memory service with semantic search |
README in short
The project's goal is to help agents learn from experience, and a diagram shows the path of knowledge: capture, curation, reuse. In the authors' internal experiment long Claude Code runs with Stash were 49% faster.