Progress Modes

OpenClaw supports two progress reporting modes to balance visibility and token cost. Choose based on your monitoring needs.

Sparse Mode (Default)

Reports only on state changes: start, milestone, error, completion.

Use this for:

Cost optimization (reduced token usage)
Tasks that run reliably without intervention
Background task queues
Default behavior when no mode specified

Example output:

[Start] PROJ-284: Implement rate limiting (via claude-code)
[Milestone] Tests passed: 45/45
[Complete] Rate limiting implemented - feature branch pushed

Trigger: No explicit request, or say:

"Start PROJ-284 in sparse mode"
"Run with sparse updates"

Live Mode

Reports state changes + periodic heartbeats (every 1-2 minutes).

Use this for:

Close monitoring of long-running tasks
Debugging stuck or slow execution
Interactive workflows requiring responsiveness
High-confidence deployments needing oversight

Example output:

[Start] PROJ-284: Implement rate limiting (via claude-code)
[Running] 2 min — compiling TypeScript...
[Running] 4 min — running tests...
[Milestone] Tests passed: 45/45
[Running] 6 min — pushing to feature branch...
[Complete] Rate limiting implemented — feature branch pushed

Trigger: Say one of:

"Start PROJ-284 with progress live"
"Run with live updates"
"Run with realtime progress"
"Run with detailed updates"

Context Compaction

Long-running sessions can accumulate token overhead. To manage context:

For Sparse Mode

Prefer sparse unless close monitoring is required
Send short milestone checkpoints instead of repeated logs
One summary per significant state change is sufficient

For Live Mode

Monitor closely but avoid reading the full log repeatedly
Use process action:log sessionId:<id> for updates instead of re-reading entire output
If the thread gets too long, post one compaction summary and continue in the same session

Best Practices

❌ Bad:  Asking for updates every 30 seconds, causing token bloat
✅ Good: Send a checkpoint once per 5-10 minute milestone

❌ Bad:  Re-reading entire 50KB execution log on each update
✅ Good: Ask for "last 100 lines of log" or new updates only

❌ Bad:  Starting new sessions for every task
✅ Good: Run multiple tasks in same session, monitor in parallel

Recommendations

Scenario	Mode	Notes
Batch processing	Sparse	Fire and forget, check results later
Interactive review	Live	Watch closely, respond to prompts
Long infrastructure tasks	Sparse	Deploy and monitor separately
Time-critical features	Live	Stay involved until complete
Development workflow	Live	Iterate quickly with feedback
CI/CD automation	Sparse	Log output sufficient

Switching Modes

You can specify progress mode in any message to OpenClaw:

"PROJ-284: Implement rate limiting" → sparse (default)
"PROJ-284: Implement rate limiting (progress: live)" → live
"Check progress for session <id> (live)" → switch to live monitoring