Claude To DeerflowSAFE

An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: claude-to-deerflow
description: "Interact with DeerFlow AI agent platform via its HTTP API. Use this skill when the user wants to send messages or questions to DeerFlow for research/analysis, start a DeerFlow conversation thread, check DeerFlow status or health, list available models/skills/agents in DeerFlow, manage DeerFlow memory, upload files to DeerFlow threads, or delegate complex research tasks to DeerFlow. Also use when the user mentions deerflow, deer flow, or wants to run a deep research task that DeerFlow can handle."
---

# DeerFlow Skill

Communicate with a running DeerFlow instance via its HTTP API. DeerFlow is an AI agent platform
built on LangGraph that orchestrates sub-agents for research, code execution, web browsing, and more.

## Architecture

DeerFlow exposes two API surfaces behind an Nginx reverse proxy:

| Service        | Direct Port | Via Proxy                        | Purpose                          |
|----------------|-------------|----------------------------------|----------------------------------|
| Gateway API    | 8001        | `$DEERFLOW_GATEWAY_URL`          | REST endpoints and embedded agent runtime |
| LangGraph-compatible API | 8001 | `$DEERFLOW_LANGGRAPH_URL`       | Agent threads, runs, streaming   |

## Environment Variables

All URLs are configurable via environment variables. **Read these env vars before making any request.**

| Variable                | Default                                  | Description                        |
|-------------------------|------------------------------------------|------------------------------------|
| `DEERFLOW_URL`          | `http://localhost:2026`                  | Unified proxy base URL             |
| `DEERFLOW_GATEWAY_URL`  | `${DEERFLOW_URL}`                        | Gateway API base (models, skills, memory, uploads) |
| `DEERFLOW_LANGGRAPH_URL`| `${DEERFLOW_URL}/api/langgraph`          | LangGraph API base (threads, runs) |

When making curl calls, always resolve the URL like this:

```bash
# Resolve base URLs from env (do this FIRST before any API call)
DEERFLOW_URL="${DEERFLOW_URL:-http://localhost:2026}"
DEERFLOW_GATEWAY_URL="${DEERFLOW_GATEWAY_URL:-$DEERFLOW_URL}"
DEERFLOW_LANGGRAPH_URL="${DEERFLOW_LANGGRAPH_URL:-$DEERFLOW_URL/api/langgraph}"
```

## Available Operations

### 1. Health Check

Verify DeerFlow is running:

```bash
curl -s "$DEERFLOW_GATEWAY_URL/health"
```

### 2. Send a Message (Streaming)

This is the primary operation. It creates a thread and streams the agent's response.

**Step 1: Create a thread**

```bash
curl -s -X POST "$DEERFLOW_LANGGRAPH_URL/threads" \
  -H "Content-Type: application/json" \
  -d '{}'
```

Response: `{"thread_id": "<uuid>", ...}`

**Step 2: Stream a run**

```bash
curl -s -N -X POST "$DEERFLOW_LANGGRAPH_URL/threads/<thread_id>/runs/stream" \
  -H "Content-Type: application/json" \
  -d '{
    "assistant_id": "lead_agent",
    "input": {
      "messages": [
        {
          "type": "human",
          "content": [{"type": "text", "text": "YOUR MESSAGE HERE"}]
        }
      ]
    },
    "stream_mode": ["values", "messages-tuple"],
    "stream_subgraphs": true,
    "config": {
      "recursion_limit": 1000
    },
    "context": {
      "thinking_enabled": true,
      "is_plan_mode": true,
      "subagent_enabled": true,
      "thread_id": "<thread_id>"
    }
  }'
```

The response is an SSE stream. Each event has the format:
```
event: <event_type>
data: <json_data>
```

Key event types:
- `metadata` — run metadata including `run_id`
- `values` — full state snapshot with `messages` array
- `messages-tuple` — incremental message updates (AI text chunks, tool calls, tool results)
- `end` — stream is complete

**Context modes** (set via `context`):
- Flash mode: `thinking_enabled: false, is_plan_mode: false, subagent_enabled: false`
- Standard mode: `thinking_enabled: true, is_plan_mode: false, subagent_enabled: false`
- Pro mode: `thinking_enabled: true, is_plan_mode: true, subagent_enabled: false`
- Ultra mode: `thinking_enabled: true, is_plan_mode: true, subagent_enabled: true`

### 3. Continue a Conversation

To send follow-up messages, reuse the same `thread_id` from step 2 and POST another run
with the new message.

### 4. List Models

```bash
curl -s "$DEERFLOW_GATEWAY_URL/api/models"
```

Returns: `{"models": [{"name": "...", "provider": "...", ...}, ...]}`

### 5. List Skills

```bash
curl -s "$DEERFLOW_GATEWAY_URL/api/skills"
```

Returns: `{"skills": [{"name": "...", "enabled": true, ...}, ...]}`

### 6. Enable/Disable a Skill

```bash
curl -s -X PUT "$DEERFLOW_GATEWAY_URL/api/skills/<skill_name>" \
  -H "Content-Type: application/json" \
  -d '{"enabled": true}'
```

### 7. List Agents

```bash
curl -s "$DEERFLOW_GATEWAY_URL/api/agents"
```

Returns: `{"agents": [{"name": "...", ...}, ...]}`

### 8. Get Memory

```bash
curl -s "$DEERFLOW_GATEWAY_URL/api/memory"
```

Returns user context, facts, and conversation history summaries.

### 9. Upload Files to a Thread

```bash
curl -s -X POST "$DEERFLOW_GATEWAY_URL/api/threads/<thread_id>/uploads" \
  -F "files=@/path/to/file.pdf"
```

Supports PDF, PPTX, XLSX, DOCX — automatically converts to Markdown.

### 10. List Uploaded Files

```bash
curl -s "$DEERFLOW_GATEWAY_URL/api/threads/<thread_id>/uploads/list"
```

### 11. Get Thread History

```bash
curl -s "$DEERFLOW_LANGGRAPH_URL/threads/<thread_id>/history"
```

### 12. List Threads

```bash
curl -s -X POST "$DEERFLOW_LANGGRAPH_URL/threads/search" \
  -H "Content-Type: application/json" \
  -d '{"limit": 20, "sort_by": "updated_at", "sort_order": "desc"}'
```

## Usage Script

For sending messages and collecting the full response, use the helper script:

```bash
bash /path/to/skills/claude-to-deerflow/scripts/chat.sh "Your question here"
```

See `scripts/chat.sh` for the implementation. The script:
1. Checks health
2. Creates a thread
3. Streams the run and collects the final AI response
4. Prints the result

## Parsing SSE Output

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

declared (5 observation(s))

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

description: "Interact with DeerFlow AI agent platform via its HTTP API. Use this skill when the user wants to send messages or questions to DeerFlow for research/analysis, start a DeerFlow conversati

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: Claude To Deerflow](https://modelsatlas.com/skills/bytedance/claude-to-deerflow/badge.svg)](https://modelsatlas.com/skills/bytedance/claude-to-deerflow/)

Not your listing, or something out of date? Claim or update your listing.