| name | notify |
| description | Inform the user what is happening — skip passive lookups |
| token_cost | 1 |
| keywords | ["inform","progress","status"] |
Inform the user
Keep the user informed of what the agent is doing.
REQUIRED: message (one-line text)
When to notify
Every step that produces effects. Editing code, running builds, checking output, fixing errors — always notify. Pure reads (find, read, ls, grep) are passive lookups and do not warrant notification.
- Before every mutating tool call (edit, bash with side effects, write)
- After every tool call that changes state or produces output worth surfacing
- On errors and unexpected output
- On completion
Message format
- message: One-line text describing the action. Max 70 chars.
- Never include IDs, codes, hashes, filenames, or specific references — keep it high-level.
{"name": "notify", "input": {"message": "Running test suite"}}
Workflow examples
Implementation task
notify("Editing config file") — before editing
notify("Running typecheck") — before running
notify("Typecheck passed") — after results
notify("Done!") — on completion
Research task
notify("Starting research") — before web searches begin
notify("Done, writing answer") — when transitioning to output
Debugging session
notify("Investigating the issue") — at start
notify("Found the cause, applying fix") — on breakthrough
notify("Fix verified, done!") — on completion
Best practices
- Skip notifications for passive lookups (find, read, ls, grep)
- Notify before mutations and after state-changing results
- Keep messages short and specific