System Prompt

View, update, and sync your agent's personality. Self-improvement across sessions.

---

Actions

Action	Description	Parameters
view	View current system prompt	—
update	Replace the system prompt	prompt, repository (optional)
add_context	Append without replacing	context
reset	Reset to default	—
get_github_context	Get GitHub event context	—

---

Basic Usage

# View current prompt
system_prompt(action="view")

# Update locally (saves to env var + .prompt file)
system_prompt(
    action="update",
    prompt="You are a specialized Python developer..."
)

# Add context without replacing
system_prompt(
    action="add_context",
    context="When working with async code, always use asyncio.gather."
)

---

GitHub Repository Sync

Persist system prompt changes across deployments by syncing to GitHub repository variables:

# Update and sync to GitHub
system_prompt(
    action="update",
    prompt="You are DevDuck, an expert AI developer...",
    repository="cagataycali/devduck"
)

# Use custom variable name
system_prompt(
    action="update",
    prompt="Specialized for data analysis...",
    repository="owner/repo",
    variable_name="DATA_AGENT_PROMPT"
)

What Gets Updated

1. ✅ Environment variable (SYSTEM_PROMPT)
2. ✅ Local .prompt file
3. ✅ GitHub repository variable (if repository specified)

---

Self-Improvement Pattern

When DevDuck discovers valuable patterns during conversations:

# Step 1: Identify new insight
insight = "Always validate JSON before parsing to avoid crashes"

# Step 2: Add to system prompt
system_prompt(action="add_context", context=insight)

# Step 3: Sync to GitHub for persistence
system_prompt(
    action="update",
    prompt=current_prompt + "\n" + insight,
    repository="cagataycali/devduck"
)

Persistent Learning

New learnings persist across sessions via the SYSTEM_PROMPT environment variable and GitHub repository variables.

---

AGENTS.md

DevDuck automatically loads AGENTS.md from the current working directory and injects it into the system prompt. This file describes project-specific context, conventions, and instructions.

# AGENTS.md — My Project

## Architecture
- FastAPI backend in /api
- React frontend in /web

## Conventions
- Use pytest for all tests
- Follow Google Python Style Guide