What is the difference between this and the standard planning skill?

The standard planning skill is adaptive and allows the AI to change course mid-task. Plan-Then-Execute is rigid and requires human approval for a fixed sequence of actions, making it safer for sensitive operations.

How does Plan-Then-Execute prevent prompt injection?

By freezing the action sequence during the planning phase, tool outputs cannot change the order or type of actions. This ensures that even if a tool returns malicious content, the AI cannot be 'tricked' into executing unintended commands.

Can I modify the plan before it runs?

Yes. The PLAN.yml file is designed for human review. You can manually edit the parameters, add or remove steps, and then set the status to 'approved' to start execution.

What happens if a step in the plan fails?

The execution logic typically defaults to a 'stop' policy on failure, ensuring that the system doesn't proceed with incorrect data. You can check the EXECUTION_LOG.md for the specific error and captured outputs.

Plan-Then-Execute

Name: Plan-Then-Execute
Author: SimHacker

bySimHacker

Security & Testing

Implements a high-integrity execution pattern that separates planning from action to prevent prompt injection and ensure human oversight.

About

The Plan-Then-Execute skill is a security-focused framework designed for high-stakes AI workflows and tasks involving untrusted data. Unlike standard adaptive agents, this skill forces Claude to generate a complete, 'frozen' sequence of actions in a YAML format which must be reviewed and approved by a human before execution begins. Once approved, the execution phase follows the plan exactly, preventing tool outputs from maliciously redirecting the agent's behavior. This pattern provides a robust audit trail and ensures that the human remains the final authority on what tools are called and in what order.

Key Features

Mandated human-in-the-loop approval gate for security compliance
Dynamic parameter resolution for data passing between sequential steps
0 GitHub stars
Frozen action sequences that prevent mid-run behavior redirection
Comprehensive execution logging for debugging and auditing
Structured YAML-based plan state management

Use Cases

Executing complex multi-step workflows that require a strict audit trail
Performing high-stakes infrastructure or database operations
Processing untrusted data from emails or web content safely

About

Key Features

Mandated human-in-the-loop approval gate for security compliance
Dynamic parameter resolution for data passing between sequential steps
0 GitHub stars
Frozen action sequences that prevent mid-run behavior redirection
Comprehensive execution logging for debugging and auditing
Structured YAML-based plan state management

Use Cases

Executing complex multi-step workflows that require a strict audit trail
Performing high-stakes infrastructure or database operations
Processing untrusted data from emails or web content safely