AntFleet

Disagreement · f5ac807a-anthropic-3

Target one-shot rule contradicts state-write description

solo Opus
repo 6f7fc663·PR #23·reviewed 1 week ago

Opus finding

Target one-shot rule contradicts state-write description

mediumbugmedium
  • skills/price-threshold-alert/SKILL.md:111-113
  • skills/price-threshold-alert/SKILL.md:181-182
Step 6 says set `announced_at=NOW` and notify subject to 4h dedup. But the invariant says `hit_at == announced_at` in practice. If the global `last_alerts.target_hit` dedup window suppresses the notification (e.g., another target hit fired 30 minutes earlier), Step 6 still sets `announced_at=NOW` regardless, which is a lie — no notification actually went out. Combined with the one-shot rule (target never re-fires once `hit_at` is set), an operator could miss a target-hit alert entirely when two targets cross close together. There is no retry mechanism.

Recommendation

Either (a) drop the 4h global dedup for target_hit (target alerts are already inherently one-shot), or (b) only set `announced_at` when the notify actually fires; leave it null and re-attempt next run if dedup suppressed.

Other reviewer

The other reviewer flagged nothing in this file/line range.

Why this didn't post

This finding didn't meet AntFleet's unanimous agreement threshold. Both frontier models review every PR independently; only findings they both flag with the same severity and category are posted to the PR. This one fell through.

read the methodology →

From the same review

These findings passed the unanimous gate on the same PR review. The disagreement above was filtered out; the findings below were posted.