Skip to content

Agent Prompt Templates

For agents to use OpenTime proactively (without being asked), add instructions to the agent's system prompt.

Usage

from opentime.prompts import get_system_prompt

# For MCP-connected agents
prompt = get_system_prompt("mcp")

# For OpenAI / Gemini function calling
prompt = get_system_prompt("openai")

# For REST API agents
prompt = get_system_prompt("rest_api", base_url="http://localhost:8080")

Then append it to your agent's system prompt:

system_prompt = f"You are a helpful coding assistant.\n\n{prompt}"

Available Modes

Mode Aliases For
"mcp" — Claude Code, Cursor, Windsurf, Cline, etc.
"function_calling" "openai", "gemini" OpenAI GPT-4, Gemini
"rest_api" "rest" Any HTTP-based agent

What the Prompt Instructs

The template tells the agent to:

  1. Always track tasks — Call task_start/task_end on every task with correlation IDs
  2. Use descriptive task types — Consistent names like "code_generation", "debugging"
  3. Check before setting timeouts — Use recommend_timeout for data-driven values
  4. Compare approaches by time — Use compare_approaches instead of guessing
  5. Monitor active tasks — Check active_tasks periodically
  6. Track everything — Not just coding, but downloads, tests, builds, searches

Manual Copy-Paste

If you can't call get_system_prompt() programmatically, here's the core text to add to any agent's instructions:

You have access to OpenTime time tracking tools. Use them proactively on EVERY task:

  1. Call task_start at the beginning of every task — save the correlation_id.
  2. Call task_end when done — pass the correlation_id.
  3. Before setting timeouts, call recommend_timeout.
  4. When choosing between approaches, call compare_approaches.
  5. Use consistent task_type names (e.g. "code_generation", "debugging", "test_run").

Do NOT wait to be asked — track time automatically on all tasks.