Skip to main content
These features are available in pflow but considered experimental. They may change, have rough edges, or be replaced in future versions. Use them if they solve a problem for you, but be aware of the caveats.

​
Built-in Git and GitHub nodes

pflow includes nodes for Git and GitHub operations, but they’re disabled by default. For most use cases, there are better options: Git operations - Use the git CLI directly via shell node. Your agent knows Git extremely well from training data and doesn’t need special nodes for common operations. 
{
  "id": "commit-changes",
  "type": "shell",
  "params": {
    "command": "git add . && git commit -m '${commit_message}'"
  }
}
GitHub operations - Use the gh CLI via shell node for simple tasks, or the GitHub MCP server for complex operations that benefit from structured tools. 
{
  "id": "create-pr",
  "type": "shell",
  "params": {
    "command": "gh pr create --title '${title}' --body '${body}'"
  }
}

​
If you still want to try the built-in nodes

Enable with: 
# Enable git nodes
pflow settings allow "pflow.nodes.git.*"

# Enable github nodes
pflow settings allow "pflow.nodes.github.*"
Then view available nodes: 
pflow registry list git
pflow registry list github

​
Natural language planner

pflow has a built-in natural language planner that lets you run commands like: 
pflow "fetch the latest issues from my repo and summarize them"
This is experimental. Using an external agent (Claude Code, Cursor, etc.) is the recommended approach - it’s more capable and provides a better experience.

​
When to use it

  • Quick one-off tasks when you’re not in an AI tool
  • Testing or exploring pflow capabilities
  • Environments where external agents aren’t available

​
Debugging planner issues

If you do use the planner and need to debug: 
pflow --trace-planner "your request"
This saves the planning process to ~/.pflow/debug/planner-trace-*.json.

​
Auto-repair

pflow can automatically attempt to fix failing workflows using LLM-powered repair. This is not enabled by default. 
pflow my-workflow --auto-repair
When enabled:
  • pflow analyzes the error
  • Attempts to fix the workflow
  • Retries execution (up to 3 attempts)

​
Caveats

  • Uses LLM calls (costs money)
  • May not fix all issues
  • Can get stuck in repair loops on unfixable problems
Your agent doesn’t know about this flag - it’s for manual experimentation.

​
Feedback

If you use these experimental features and have feedback: