Openharness Daily: The Prompt Miner Finally Mines Something
A cluster of four fixes closed out a nineteen-day streak of failed prompt-miner runs, while a new release pipeline shipped and immediately needed a documentation follow-up. Both threads share the same lesson: fixes that only look correct on the happy path keep coming back.
Duration: PT2M31S
Episode overview
This episode is a short developer briefing from Openharness Daily.
It explains recent repository work in plain language.
- Show: Openharness Daily
- Published: 2026-08-03T12:00:34Z
- Audio duration: PT2M31S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's August third, and today's activity is dominated by one story: getting the prompt-miner cron to actually work, after it never once succeeded.
The headline number is stark. Before today, the prompt-miner's lifetime record was nineteen NO-CORPUS results and four NO-SESSIONS, zero completions. PR 698 explains why: subagent traces were being counted as human turns, poisoning the scoring engine, and the mining window was too narrow to ever hit the required…
That fix sits alongside three others in the same family. PR 695 found the miner's entry point silently exited with zero output when invoked through its documented symlink path — a bug that survived twenty-three days because the existing tests hardcoded the real path and never exercised the symlink. PR 696 found…
The common thread: each of these bugs was invisible to its own test suite, and each one shipped a second, smaller fix within hours after adversarial review caught an edge case the first pass missed. That pattern is worth remembering — the review step is catching real regressions here, not just style nits.
Second theme: release infrastructure. PR 690 replaced tag-triggered publishing with releasing every…
What's…