Adhoc

Ad-hoc Prompts

Ad-hoc prompts let you run a single AI prompt against a workspace without starting a full agentic session. Useful for quick tasks, code reviews, or one-off questions.

Start an Ad-hoc Prompt

Endpoint: POST /api/v1/workspaces/{name}/adhoc

curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Review the authentication code and identify any security issues",
    "model": "anthropic/claude-opus-4-6"
  }'

Request:

{
  "prompt": "Review the authentication code and identify any security issues",
  "model": "anthropic/claude-opus-4-6"
}

Response:

{
  "running": true,
  "output": "",
  "message": "ad-hoc prompt started"
}

If already running:

{
  "running": true,
  "output": "$ opencode run -m anthropic/claude-opus-4-6...\nprompt: Review the...",
  "message": "ad-hoc prompt already running"
}

Model Format

The model parameter accepts the same format as GOAL.md models:

• "anthropic/claude-opus-4-6" — standard model
• "anthropic/claude-opus-4-6 (max)" — model with variant
• Use GET /api/v1/models to list available models

How Ad-hoc Works

Ad-hoc prompts run opencode run -m {model} --agent build --title "adhoc [{model}]" with your prompt piped as stdin. The workspace directory is used as the working directory.

Get Ad-hoc Status

Check if an ad-hoc prompt is running and get its current output.

Endpoint: GET /api/v1/workspaces/{name}/adhoc

curl -s $BASE_URL/api/v1/workspaces/my-project/adhoc

Response (running):

{
  "running": true,
  "output": "$ opencode run -m anthropic/claude-opus-4-6 --agent build --title \"adhoc [anthropic/claude-opus-4-6]\"\nprompt: Review the authentication code...\n\nAnalyzing the codebase...\n\nFound 3 potential issues:\n1. JWT tokens lack expiration...",
  "message": "adhoc status"
}

Response (not running):

{
  "running": false,
  "output": "$ opencode run...\n...\n[completed successfully]",
  "message": "adhoc status"
}

Polling for Completion

WORKSPACE="my-project"

# Start the prompt
curl -X POST $BASE_URL/api/v1/workspaces/$WORKSPACE/adhoc \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Summarize the GOAL.md", "model": "anthropic/claude-sonnet-4-6"}'

# Poll until done
while true; do
  STATUS=$(curl -s $BASE_URL/api/v1/workspaces/$WORKSPACE/adhoc)
  RUNNING=$(echo $STATUS | jq '.running')
  
  if [ "$RUNNING" = "false" ]; then
    echo "Ad-hoc complete!"
    echo $STATUS | jq -r '.output'
    break
  fi
  
  echo "Still running..."
  sleep 3
done

Stop an Ad-hoc Prompt

Stop a running ad-hoc prompt.

Endpoint: DELETE /api/v1/workspaces/{name}/adhoc

curl -X DELETE $BASE_URL/api/v1/workspaces/my-project/adhoc

Response:

{
  "running": false,
  "output": "$ opencode run...\n...\n[process terminated]",
  "message": "ad-hoc stopped"
}

Ad-hoc vs Full Session

Common Ad-hoc Use Cases

# Code review
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Review all Go files for potential race conditions", "model": "anthropic/claude-opus-4-6"}'

# Documentation generation
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Generate a README.md for this project", "model": "anthropic/claude-sonnet-4-6"}'

# Quick fix
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "Fix the failing test in auth_test.go", "model": "anthropic/claude-sonnet-4-6"}'

# Analysis
curl -X POST $BASE_URL/api/v1/workspaces/my-project/adhoc \
  -d '{"prompt": "List all TODO comments in the codebase", "model": "anthropic/claude-haiku-4-5"}'