openclaw - 💡(How to fix) Fix Cron classifier can mark successful runs as error based on denial-token substring matching in the summary

Summary

Cron status classification appears to scan the final natural-language summary for denial tokens. This can produce false errors when a successful run includes a benign phrase such as did not run in a non-failure context.

Actual behavior

A run completed successfully, wrote its expected outputs, and explicitly reported success, but the recorded cron status was still error because the summary also included a sentence saying:

• I did not run a separate HTML validator in this cron turn.

That benign verification note triggered the denial-token classifier even though the job itself had succeeded.

Expected behavior

A successful run should not be marked error based on substring matching inside a free-form summary. Outcome classification should use structured signals, not incidental wording.

Why this matters

Natural-language summaries routinely contain phrases like:

• did not run
• not requested
• did not need to
• was not required

Those phrases are too ambiguous to drive final job status. The result is false negatives that make operators distrust the cron ledger.

Minimal repro

1. Run a skill successfully so that expected outputs are written
2. Include a benign sentence in the final summary such as I did not run a separate validator
3. Observe the cron run being classified as error because of denial-token matching

Expected fix direction

• derive final status from explicit structured outcome fields
• treat the summary as human-readable explanation only
• avoid substring-based denial-token classification for final success or failure

Environment

• OpenClaw 2026.5.19
• Linux