Draft GitHub Issue

Title: Cron session accumulation can silently exhaust agent context window — no guardrails or warnings

Problem

High-frequency OpenClaw cron jobs (e.g. every 15m / every 30m) continuously create isolated run sessions and grow the session store without visible guardrails. In a real incident, the main agent session context grew to 152k/272k (56%) and the session store reached 113 active entries. Eventually, the agent stopped replying to new inbound messages even though channels looked healthy (“all green”) — the logs showed dispatch succeeded but produced replies=0.

This failure mode is hard to diagnose because gateway/channel status can appear normal while the agent silently fails due to context/session pressure.

Evidence / Repro signals (from incident)

• In openclaw logs --follow (and likely /tmp/openclaw/openclaw-2026-03-07.log), messages were dispatched but no reply produced:
  • [feishu] feishu[default]: dispatching to agent (session=agent:main:main)
  • [feishu] feishu[default]: dispatch complete (queuedFinal=false, replies=0)
• openclaw sessions showed:
  • agent:main:main direct ... gpt-5.2 152k/272k (56%)
• openclaw sessions cleanup did not reduce entries:
  • Applied maintenance. Current entries: 113 (before/after unchanged)

Root causes (missing guardrails)

1. Cron run sessions have no TTL / auto-expiry, so store grows indefinitely
2. No session count / context usage threshold warnings
3. No guardrail on creating high-frequency cron jobs (no warning, no recommended minimum interval)
4. sessions cleanup appears too weak for completed cron runs (does not reclaim)

Impact

Any user using cron can accidentally degrade the assistant until it becomes non-responsive, with a “green status but not working” symptom.

Environment

• OpenClaw 2026.3.2
• macOS
• ~8 cron jobs; high-frequency examples:
  • fa44760a-b511-42bd-aeec-76233a515fa5 — watch-openclaw-issue-pr-37232-37234 — every 15m
  • d5b80f10-6020-4cec-8689-f97fdc68530c — watch-openai-codex-gpt54 — every 30m

Workaround

Disable high-frequency cron jobs + restart gateway:

• openclaw cron disable <jobId>
• openclaw gateway restart

Proposed solutions

• Session TTL for cron runs: completed isolated cron run sessions auto-expire (e.g. 24h) and are reclaimed
• Context usage warnings: warn when main session context exceeds threshold (e.g. 70%)
• Cron frequency guardrail: warn when creating jobs with interval < 30m
• Improve sessions cleanup: actually reclaim completed cron run sessions