agentation-watch-mode

Agentation watch mode

When to use: The user says watch mode, wants continuous Agentation feedback handling, or asks you to keep watching annotations and applying fixes as new feedback arrives.

Goal

Stay in a watch loop so the user can annotate in the browser and have those annotations processed continuously.

Workflow

1. Start a watch loop with agentation_agentation_watch_annotations.
2. For each returned annotation:
   • Call agentation_agentation_acknowledge immediately so the user can see you picked it up.
   • Read any extra session context you need.
   • Make the requested fix.
   • Call agentation_agentation_resolve with a short summary of what changed.
3. After the batch is done, call agentation_agentation_watch_annotations again.
4. Continue until the user says stop or the watch call times out.

Defaults

• If the user gives a session ID, watch that session.
• If they do not, watch all sessions.
• Use a short batching window so nearby annotations are grouped together.
• Use a reasonable timeout and report when it expires.

Handling ambiguous feedback

• If an annotation is unclear, reply on the thread with a short clarifying question before making speculative changes.
• If you cannot complete the fix safely, leave the annotation open and explain the blocker in a reply.

Response style

• Keep progress updates brief.
• In each resolve summary, state what you changed, not just that it is done.