Summary

Add an operator-friendly live progress view/log source that shows active sessions, tool calls, tool results, model completion, and run status without requiring users to manually parse per-session trajectory JSONL files.

Problem to solve

In OpenClaw 2026.5.12, detailed execution activity is present in per-session trajectory files under ~/.openclaw/agents/main/sessions/*.trajectory.jsonl, but the old habit of monitoring ~/.openclaw/logs no longer answers the operational question: what is OpenClaw doing right now, and did the step succeed?

For heavy operators running OpenClaw in long-lived chat/gateway setups, this makes live diagnosis harder. ~/.openclaw/logs now mainly contains config audit, health, and gateway stability/failure information. It is useful for startup/config issues, but not for normal agent progress. The useful events exist, but are structured and scattered by session file.

Proposed solution

Provide a first-class human-readable progress surface, for example one or more of:

• openclaw logs --source sessions --follow or openclaw logs --source trajectory --follow
• openclaw sessions tail --follow
• a compact live view in openclaw status / TUI showing recent active session events
• an option to render trajectory JSONL as concise lines with timestamp, session key, event type, model/provider, tool name, success/error/timeout status, and a short redacted argument/result preview when safe

The important behavior is that operators should not need to know the latest trajectory filename or hand-write jq to monitor active work.

Alternatives considered

Manual workaround today:

latest=$(find ~/.openclaw/agents/main/sessions -maxdepth 1 -name '*.trajectory.jsonl' -printf '%T@ %p\n' | sort -n | tail -1 | cut -d' ' -f2-)
tail -f "$latest" | jq -r '[.ts,.type,(.sessionKey // ""),(.data.name // .data.status // .data.provider // .data.modelId // "")] | @tsv'

This works, but it is fragile for multiple concurrent sessions, is not discoverable, and exposes raw event structure instead of an operator-focused view.

openclaw gateway stability and ~/.openclaw/logs/stability are useful for process stability diagnostics, but they do not replace live task-progress monitoring.

Impact

Affected: Operators running OpenClaw as a long-lived gateway/chat automation service.

Severity: Medium. It does not block execution, but it makes incident response and long-running task supervision much harder.

Frequency: Always, for users who previously relied on tailing runtime logs to see what OpenClaw was executing.

Consequence: Extra manual work, slower debugging, and uncertainty about whether OpenClaw is currently thinking, calling tools, waiting, failing, or already finished.

Evidence/examples

Observed on OpenClaw 2026.5.12 (f066dd2) on Debian Linux.

Current ~/.openclaw/logs contents are config/stability oriented, for example:

~/.openclaw/logs/config-audit.jsonl
~/.openclaw/logs/config-health.json
~/.openclaw/logs/stability/openclaw-stability-*.json

Detailed runtime events are present in:

~/.openclaw/agents/main/sessions/*.trajectory.jsonl

Observed trajectory event types include:

session.started
context.compiled
prompt.submitted
tool.call
tool.result
model.completed
session.ended

This is valuable data, but not a convenient live operations log.

Additional information

Related but not identical existing issue: #66841 covers stale/misleading openclaw logs --follow output after side-by-side cutover. This request is specifically about a human-readable live task/session progress view over the current trajectory/session logging model.