Run Logs

Every run is recorded on disk so it can be inspected and replayed. Logs are always finalized, even when a run fails.

Run directory

Each run creates a directory under ~/.team-harness/runs/<run-id>/ containing:

• run.json — the full delta-based run log, a losslessly replayable record of the conversation.
• todo.json — the persistent task list (written once the coordinator first calls todo_write, so it may be absent for runs that never used a task list).

View a run log from the CLI:

th logs            # list recent runs
th logs <run-id>   # print one run's full log

Worker session manifest

Each run also writes <output_dir>/<run-id>/worker_sessions.json — a compact index of every spawned worker with its prompt, status, timestamps, log paths, and resume-related metadata. (output_dir comes from [coordinator].output_dir; see Configuration.)

Per-worker logs

Each worker's stdout and stderr are written under the same session output directory:

<output_dir>/<run-id>/workers/<worker-label>__<agent-id>/stdout.jsonl
<output_dir>/<run-id>/workers/<worker-label>__<agent-id>/stderr.log

When the coordinator supplies a worker_label, it is treated as a filesystem-safe label, not a path — team-harness rejects labels containing path separators and keeps all worker logs inside the session output directory. Workers spawned without a label use their generated agent id as the directory name.

Retry and failure records

Transient coordinator API failures (429 / 5xx) are recorded in run.json under coordinator_retries, and the terminal run-level failure under failure. Retry behavior is tuned with max_retries, retry_base_delay_s, and retry_max_delay_s — see Configuration.