task-progress-stream
Use this skill when the user wants real-time progress updates inside the OpenClaw chat UI for a long-running task such as:
- model training
- evaluation
- inference
- data preprocessing
- long shell jobs
- benchmark runs
It works in two modes:
-
run
Start a command, capture stdout/stderr, parse common metrics, and periodically inject progress summaries into chat. -
tail
Monitor an existing log file and periodically inject parsed progress summaries into chat.
What it extracts
It tries to recognize common fields from logs:
- epoch / max_epoch
- step / max_step
- loss
- learning rate
- validation metric
- best metric
- ETA
- speed
Typical usage
Start a training job and stream progress into chat
node ./scripts/task_progress_stream.js run \
--session main \
--label train \
--cwd /path/to/project \
--cmd "python src/train.py --config configs/train.yaml" \
--interval-sec 20