MCPInstall

deep-agents-memory

INVOKE THIS SKILL when your Deep Agent needs memory, persistence, or filesystem access. Covers StateBackend (ephemeral), StoreBackend (persistent), FilesystemMiddleware, and CompositeBackend for routing.

Installs
55
Last 24h
—
Trending rank
—
Security
Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/langchain-ai/skills-benchmarks --skill deep-agents-memory
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads deep-agents-memory automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

<overview> Deep Agents use pluggable backends for file operations and memory:

Short-term (StateBackend): Persists within a single thread, lost when thread ends Long-term (StoreBackend): Persists across threads and sessions Hybrid (CompositeBackend): Route different paths to different backends

FilesystemMiddleware provides tools: ls, read_file, write_file, edit_file, glob, grep </overview>

<backend-selection>
Use CaseBackendWhy
Temporary working filesStateBackendDefault, no setup
Local development CLIFilesystemBackendDirect disk access
Cross-session memoryStoreBackendPersists across threads
Hybrid storageCompositeBackendMix ephemeral + persistent
</backend-selection> <ex-default-state-backend> <python> Default StateBackend stores files ephemerally within a thread. ```python from deepagents import create_deep_agent

agent = create_deep_agent() # Default: StateBackend result = agent.invoke({ "messages": [{"role": "user", "content": "Write notes to /draft.txt"}] }, config={"configurable": {"thread_id": "thread-1"}})

/draft.txt is lost when thread ends

</python>
<typescript>
Default StateBackend stores files ephemerally within a thread.
```typescript
import { createDeepAgent } from "deepagents";

const agent = await createDeepAgent();  // Default: StateBackend
const result = await agent.invoke({
  messages: [{ role: "user", content: "Write notes to /draft.txt" }]
}, { configurable: { thread_id: "thread-1" } });
// /draft.txt is lost when thread ends
</typescript> </ex-default-state-backend> <ex-composite-backend-for-hybrid> <python> Configure CompositeBackend to route paths to different storage backends. ```python from deepagents import create_deep_agent from deepagents.backends import CompositeBackend, StateBackend, StoreBackend from langgraph.store.memory import InMemoryStore

store = InMemoryStore()

composite_backend = lambda rt: CompositeBackend( default=StateBackend(rt), routes={"/memories/": StoreBackend(rt)} )

agent = create_deep_agent(backend=composite_backend, store=store)

/draft.txt -> ephemeral (StateBackend)

/memories/user-prefs.txt -> persistent (StoreBackend)

</python>
<typescript>
Configure CompositeBackend to route paths to different storage backends.
```typescript
import { createDeepAgent, CompositeBackend, StateBackend, StoreBackend } from "deepagents";
import { InMemoryStore } from "@langchain/langgraph";

const store = new InMemoryStore();

const agent = await createDeepAgent({
  backend: (config) => new CompositeBackend(
    new StateBackend(config),
    { "/memories/": new StoreBackend(config) }
  ),
  store
});

// /draft.txt -> ephemeral (StateBackend)
// /memories/user-prefs.txt -> persistent (StoreBackend)
</typescript> </ex-composite-backend-for-hybrid> <ex-cross-session-memory> <python> Files in /memories/ persist across threads via StoreBackend routing. ```python # Using CompositeBackend from previous example config1 = {"configurable": {"thread_id": "thread-1"}} agent.invoke({"messages": [{"role": "user", "content": "Save to /memories/style.txt"}]}, config=config1)

config2 = {"configurable": {"thread_id": "thread-2"}} agent.invoke({"messages": [{"role": "user", "content": "Read /memories/style.txt"}]}, config=config2)

Thread 2 can read file saved by Thread 1

</python>
<typescript>
Files in /memories/ persist across threads via StoreBackend routing.
```typescript
// Using CompositeBackend from previous example
const config1 = { configurable: { thread_id: "thread-1" } };
await agent.invoke({ messages: [{ role: "user", content: "Save to /memories/style.txt" }] }, config1);

const config2 = { configurable: { thread_id: "thread-2" } };
await agent.invoke({ messages: [{ role: "user", content: "Read /memories/style.txt" }] }, config2);
// Thread 2 can read file saved by Thread 1
</typescript> </ex-cross-session-memory> <ex-filesystem-backend-local-dev> <python> Use FilesystemBackend for local development with real disk access and human-in-the-loop. ```python from deepagents import create_deep_agent from deepagents.backends import FilesystemBackend from langgraph.checkpoint.memory import MemorySaver

agent = create_deep_agent( backend=FilesystemBackend(root_dir=".", virtual_mode=True), # Restrict access interrupt_on={"write_file": True, "edit_file": True}, checkpointer=MemorySaver() )

Agent can read/write actual files on disk

</python>
<typescript>
Use FilesystemBackend for local development with real disk access and human-in-the-loop.
```typescript
import { createDeepAgent, FilesystemBackend } from "deepagents";
import { MemorySaver } from "@langchain/langgraph";

const agent = await createDeepAgent({
  backend: new FilesystemBackend({ rootDir: ".", virtualMode: true }),
  interruptOn: { write_file: true, edit_file: true },
  checkpointer: new MemorySaver()
});
</typescript>

Security: Never use FilesystemBackend in web servers - use StateBackend or sandbox instead. </ex-filesystem-backend-local-dev>

<ex-store-in-custom-tools> <python> Access the store directly in custom tools for long-term memory operations. ```python from langchain.tools import tool, ToolRuntime from langchain.agents import create_agent from langgraph.store.memory import InMemoryStore

@tool def get_user_preference(key: str, runtime: ToolRuntime) -> str: """Get a user preference from long-term storage.""" store = runtime.store result = store.get(("user_prefs",), key) return str(result.value) if result else "Not found"

@tool def save_user_preference(key: str, value: str, runtime: ToolRuntime) -> str: """Save a user preference to long-term storage.""" store = runtime.store store.put(("user_prefs",), key, {"value": value}) return f"Saved {key}={value}"

store = InMemoryStore()

agent = create_agent( model="gpt-4.1", tools=[get_user_preference, save_user_preference], store=store )

</python>
</ex-store-in-custom-tools>

<boundaries>
### What Agents CAN Configure

- Backend type and configuration
- Routing rules for CompositeBackend
- Root directory for FilesystemBackend
- Human-in-the-loop for file operations

### What Agents CANNOT Configure

- Tool names (ls, read_file, write_file, edit_file, glob, grep)
- Access files outside virtual_mode restrictions
- Cross-thread file access without proper backend setup
</boundaries>

<fix-storebackend-requires-store>
<python>
StoreBackend requires a store instance.
```python
# WRONG
agent = create_deep_agent(backend=lambda rt: StoreBackend(rt))

# CORRECT
agent = create_deep_agent(backend=lambda rt: StoreBackend(rt), store=InMemoryStore())
</python> <typescript> StoreBackend requires a store instance. ```typescript // WRONG const agent = await createDeepAgent({ backend: (c) => new StoreBackend(c) });

// CORRECT const agent = await createDeepAgent({ backend: (c) => new StoreBackend(c), store: new InMemoryStore() });

</typescript>
</fix-storebackend-requires-store>

<fix-statebackend-files-dont-persist>
<python>
StateBackend files are thread-scoped - use same thread_id or StoreBackend for cross-thread access.
```python
# WRONG: thread-2 can't read file from thread-1
agent.invoke({"messages": [...]}, config={"configurable": {"thread_id": "thread-1"}})  # Write
agent.invoke({"messages": [...]}, config={"configurable": {"thread_id": "thread-2"}})  # File not found!
</python> <typescript> StateBackend files are thread-scoped - use same thread_id or StoreBackend for cross-thread access. ```typescript // WRONG: thread-2 can't read file from thread-1 await agent.invoke({ messages: [...] }, { configurable: { thread_id: "thread-1" } }); // Write await agent.invoke({ messages: [...] }, { configurable: { thread_id: "thread-2" } }); // File not found! ``` </typescript> </fix-statebackend-files-dont-persist> <fix-path-prefix-for-persistence> <python> Path must match CompositeBackend route prefix for persistence. ```python # With routes={"/memories/": StoreBackend(rt)}: agent.invoke(...) # /prefs.txt -> ephemeral (no match) agent.invoke(...) # /memories/prefs.txt -> persistent (matches route) ``` </python> <typescript> Path must match CompositeBackend route prefix for persistence. ```typescript // With routes: { "/memories/": StoreBackend }: await agent.invoke(...); // /prefs.txt -> ephemeral (no match) await agent.invoke(...); // /memories/prefs.txt -> persistent (matches route) ``` </typescript> </fix-path-prefix-for-persistence> <fix-production-store> <python> Use PostgresStore for production (InMemoryStore lost on restart). ```python # WRONG # CORRECT store = InMemoryStore() store = PostgresStore(connection_string="postgresql://...") ``` </python> <typescript> Use PostgresStore for production (InMemoryStore lost on restart). ```typescript // WRONG // CORRECT const store = new InMemoryStore(); const store = new PostgresStore({ connectionString: "..." }); ``` </typescript> </fix-production-store> <fix-filesystem-backend-needs-virtual-mode> <python> Enable virtual_mode=True to restrict path access (prevents ../ and ~/ escapes). ```python backend = FilesystemBackend(root_dir="/project", virtual_mode=True) # Secure ``` </python> </fix-filesystem-backend-needs-virtual-mode> <fix-longest-prefix-match> <python> CompositeBackend matches longest prefix first. ```python routes = {"/mem/": StoreBackend(rt), "/mem/temp/": StateBackend(rt)} # /mem/file.txt -> StoreBackend, /mem/temp/file.txt -> StateBackend (longer match) ``` </python> </fix-longest-prefix-match>

FAQ

What does the deep-agents-memory skill do?

INVOKE THIS SKILL when your Deep Agent needs memory, persistence, or filesystem access. Covers StateBackend (ephemeral), StoreBackend (persistent), FilesystemMiddleware, and CompositeBackend for routing.

How do I install deep-agents-memory?

Run npx skills add https://github.com/langchain-ai/skills-benchmarks --skill deep-agents-memory in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/deep-agents-memory/ (personal) or .claude/skills/deep-agents-memory/ (project).

Which AI agents can use deep-agents-memory?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is deep-agents-memory safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains deep-agents-memory?

It is published by langchain-ai in langchain-ai/skills-benchmarks, and it is an official skill from the maker of the technology it covers. It has 55 recorded installs.

More from langchain-ai/skills-benchmarks

View all →

react-components

langchain-ai/skills-benchmarks

Modern React component patterns with hooks and TypeScript

OfficialFrontend & Design85 installs

langgraph-fundamentals

langchain-ai/skills-benchmarks

Open the skill to read its SKILL.md instructions.

OfficialAI & Agents64 installs

langsmith-trace

langchain-ai/skills-benchmarks

Open the skill to read its SKILL.md instructions.

OfficialAI & Agents64 installs

langchain-oss-primer

langchain-ai/skills-benchmarks

Open the skill to read its SKILL.md instructions.

OfficialAI & Agents62 installs

langchain-dependencies

langchain-ai/skills-benchmarks

Open the skill to read its SKILL.md instructions.

OfficialAI & Agents61 installs

deep-agents-core

langchain-ai/skills-benchmarks

Open the skill to read its SKILL.md instructions.

OfficialAI & Agents60 installs

Popular ai & agents skills

View category →

google-agents-cli-publish

google/agents-cli

This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).

AI & Agents513.9K installs

google-agents-cli-eval

google/agents-cli

This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).

AI & Agents513.5K installs

skill-creator

anthropics/skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

OfficialAI & Agents416.4K installs

codex-pet

prime-skills/runcomfy-agent-skills

Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.

AI & Agents402.1K installs

full-output-enforcement

leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

AI & Agents401K installs

executing-plans

obra/superpowers

Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available

AI & Agents245.7K installs