Custom System Prompt

payload = {
    "system_prompt": "You are a research assistant...",
    "initial_message": {
        "role": "user",
        "content": [{"type": "text", "text": "Say hello."}],
    },
    "title": "custom-system-prompt demo",
}
task = requests.post(
    f"{base_url}/api/v1/app-conversations",
    headers={"Authorization": f"Bearer {api_key}"},
    json=payload,
).json()
task_id = task["id"]

while not conv_id:
    time.sleep(3)
    item = requests.get(
        f"{base_url}/api/v1/app-conversations/start-tasks",
        headers=headers,
        params={"ids": task_id},
    ).json()[0]
    if item["status"] == "ERROR":
        raise RuntimeError(item.get("detail"))
    conv_id = item.get("app_conversation_id")
    sandbox_id = item.get("sandbox_id") or sandbox_id

event = requests.get(
    f"{base_url}/api/v1/conversation/{conv_id}/events/search",
    headers=headers,
    params={"kind__eq": "SystemPromptEvent", "limit": 1},
).json()["items"][0]

assert custom_prompt in event["system_prompt"]["text"]

requests.delete(f"{base_url}/api/v1/app-conversations/{conv_id}", headers=headers)
requests.delete(
    f"{base_url}/api/v1/sandboxes/{sandbox_id}",
    headers=headers,
    params={"sandbox_id": sandbox_id},
)

pip install requests
export OH_API_KEY="your-api-key"   # Cloud: Profile → API Keys.
                                   # OHE: Settings → API Keys on your instance.

python custom_system_prompt.py

# or against an OpenHands Enterprise instance:
python custom_system_prompt.py --base-url https://your-ohe.example.com

# supply your own prompt / message:
python custom_system_prompt.py \
    --prompt "You are Sir Reginald, a pirate. Every reply starts with 'Arrr!'" \
    --message "Say hello."

# keep resources for inspection (no cleanup at the end):
python custom_system_prompt.py --keep

=== start conversation ===
  custom prompt: You are a research assistant. Answer questions concisely and accurately. When wr...
  start-task status: WAITING_FOR_SANDBOX
  start-task status: WAITING_FOR_SANDBOX
  start-task status: WAITING_FOR_SANDBOX
  start-task status: WAITING_FOR_SANDBOX
  start-task status: SETTING_UP_SKILLS
  start-task status: SETTING_UP_SKILLS
  start-task status: STARTING_CONVERSATION
  start-task status: STARTING_CONVERSATION
  start-task status: READY
conversation: 7698ab13bb994b3181ea01d23652a2ca
sandbox: 3U13tOWV00GZpo1bhnhMBf

=== verify SystemPromptEvent ===
  system_prompt.text (124 chars): You are a research assistant. Answer questions concisely and accurately. When writing code, focus on clarity and simplic...
  dynamic_context.text (47806 chars) — appended automatically; contains repo/skills/secrets/datetime blocks

  PASS: custom prompt found in SystemPromptEvent

=== cleanup ===
  deleted conversation 7698ab13bb994b3181ea01d23652a2ca
  deleted sandbox 3U13tOWV00GZpo1bhnhMBf

You are a research assistant. Answer questions concisely and accurately.
When writing code, focus on clarity and simplicity. Always cite your sources
when making factual claims.

You are a code reviewer. Analyze code for correctness, security, and
maintainability. Provide specific, actionable feedback. Focus on material
issues, not style preferences.

You are a technical documentation writer. Write clear, accurate, and
comprehensive documentation. Use examples to illustrate concepts.
Structure content with clear headings and logical flow.

You are a data analyst. Analyze data to extract insights and patterns.
Use appropriate statistical methods and visualizations. Explain your
findings clearly with supporting evidence.

payload = {
    "system_prompt": "You are a research assistant...",
    "system_message_suffix": "Focus on academic papers published after 2020.",
    ...
}

[system_prompt.text          — your custom static prompt]

[dynamic_context.text        — repo context, skills, secrets, datetime]
[system_message_suffix       — appended to the dynamic context]

curl -s https://your-deployment/openapi.json \
  | jq '.components.schemas["AppConversationStartRequest-Input"].properties.system_prompt'