checkpoint-format
Interactive checkpoint format for orchestrator pause points
When & Why to Use This Skill
This Claude skill provides a standardized, interactive checkpoint format designed for AI orchestrators to manage pause points and user approvals effectively. By implementing structured templates for state summaries, error recovery, and depth limit management, it ensures that complex agentic workflows remain transparent, controllable, and reliably integrated with human-in-the-loop decision-making.
Use Cases
- Critical Path Approvals: Implementing mandatory pause points in workflows like 'pr-review-manager' or 'architecture-lead' to get user sign-off before proceeding to implementation or merging.
- Automated Error Recovery: Providing a structured interface when a sub-agent fails, allowing users to choose between retrying the task, skipping the step, or launching a debug analyst.
- Complexity Management: Handling depth limit triggers in recursive agentic tasks by prompting the user to either increase limits or simplify the task decomposition.
- State Synchronization: Using the standard checkpoint template to provide clear summaries of current progress and primary action options during long-running multi-stage operations.
| name | checkpoint-format |
|---|---|
| description | Interactive checkpoint format for orchestrator pause points |
Checkpoint Format Skill
Use this format when orchestrators need to pause for user approval.
Standard Checkpoint Template
★ CHECKPOINT: [Name]
[Summary of current state]
Options:
- A) [Primary action]
- B) [Alternative]
- C) [Abort]
IMPORTANT: Wait for user to select an option before proceeding.
Error Recovery Checkpoint
When a sub-agent fails:
★ CHECKPOINT: Sub-Agent Error
**Failed Agent:** {agent_name}
**Task:** {task_description}
**Error:** {error_message}
### Partial Progress
{list of completed steps before failure}
### Options
- **R) Retry** - Attempt {agent_name} again
- **S) Skip** - Continue to next step (may cause issues)
- **A) Abort** - Stop workflow, state saved to 00-index.md
- **D) Debug** - Launch debug-analyst to investigate
**[WAIT FOR USER INPUT]**
Depth Limit Checkpoint
When spawning depth limit is reached:
★ CHECKPOINT: Depth Limit Reached
**Current Depth:** 5/5
**Attempted Spawn:** {agent_name}
**Reason:** {why spawn was needed}
This may indicate a recursive workflow or overly complex task.
### Options
- **I) Increase** - Raise limit to 7 and continue
- **S) Simplify** - Break task into smaller pieces
- **A) Abort** - Stop and review workflow design
**[WAIT FOR USER INPUT]**
Orchestrator-Specific Checkpoints
| Orchestrator | Checkpoint Points |
|---|---|
pr-review-manager |
After review, Before implement, Before merge |
architecture-lead |
After analysis, After design options |
implementation-manager |
After decomposition, Before each task |