Summary

Add support for awaiting_input todo status to allow agents to signal intentional pauses while waiting for user input at interactive checkpoints.

Problem

The todo-continuation-enforcer is excellent for ensuring agents complete their work—it's one of the key features that makes Sisyphus so effective. However, some agent workflows intentionally require user input at specific checkpoints:

• Discovery agents: Need user confirmation on project-specific configuration before proceeding
• Planning agents: Present a plan and wait for approval before execution
• Review workflows: Show findings and ask "should I proceed?" or "anything to correct?"
• Multi-phase workflows: Pause between phases for user checkpoint

Currently, when an agent pauses to ask for user input while having pending todos, the enforcer interprets this as "agent stopped prematurely" and injects continuation prompts—preventing the user from interacting.

Solution

Add awaiting_input as a recognized todo status that signals "this task is intentionally paused waiting for external input."

This is a minimal, semantic change that:

• Lets agents explicitly signal when they're waiting for user input
• Preserves auto-continuation for all other scenarios
• Requires no configuration changes
• Works for any agent without pre-registration

Changes

// Before
function getIncompleteCount(todos: Todo[]): number {
  return todos.filter(t => t.status !== "completed" && t.status !== "cancelled").length
}

// After  
function getIncompleteCount(todos: Todo[]): number {
  return todos.filter(t => 
    t.status !== "completed" && 
    t.status !== "cancelled" &&
    t.status !== "awaiting_input"
  ).length
}

Also added:

• getAwaitingInputCount() helper function
• Logging when tasks are awaiting user input

Example Usage

When an agent needs user input:

1. Mark the interactive task as awaiting_input
2. Ask the user your question
3. After user responds, mark the task in_progress or completed and continue

Agent todos:
  1. [completed] Scan codebase
  2. [completed] Detect technology stack  
  3. [awaiting_input] Define hard rules    ← Waiting for user
  4. [pending] Generate documentation

Hook behavior:
  - Sees awaiting_input task
  - Recognizes intentional pause
  - Does NOT inject continuation

Why This Approach

Backward Compatibility

• Fully backward compatible—existing agents/workflows unchanged
• Agents that don't use awaiting_input behave exactly as before
• No config schema changes required
• No breaking changes to existing behavior

Testing

The existing test file should be extended with:

• Test that continuation is skipped when todos have awaiting_input status
• Test that continuation resumes when awaiting_input task is changed to in_progress

Thank you for building such a powerful agent framework—the todo-continuation-enforcer is genuinely one of the features that makes oh-my-opencode stand out. This small addition would make it even more flexible for diverse agent workflows.

Summary by cubic

Add support for the awaiting_input todo status so agents can pause at interactive checkpoints without being forced to continue. This keeps normal auto-continuation for all other cases.

• New Features
  • Exclude awaiting_input from the incomplete todo count.
  • Add getAwaitingInputCount() and log when tasks are waiting for input.

^{Written for commit 8ca8ace. Summary will update on new commits.}