AntFleet

Disagreement · be00b9c6-anthropic-0

Heartbeat exclusion in DEFAULT_FLIP_DISABLE contradicts inheritance rule

mismatch
repo 6f7fc663·PR #5·reviewed 1 week ago

Primary finding

Heartbeat exclusion in DEFAULT_FLIP_DISABLE contradicts inheritance rule

mediumbughigh
  • skills/fork-skill-digest/SKILL.md:92-96
  • skills/fork-skill-digest/SKILL.md:281
The constraint justifying the heartbeat exclusion claims that 'every fork inheriting upstream's enabled: true would game the disable count if any fork explicitly sets enabled: false to silence it.' But step 6 explicitly defines forks_disabled_count to only include forks that EXPLICITLY set enabled:false (not inherited). So the gaming scenario described is the intended, correct counting behavior, not a bug. The justification comment is misleading and the special-case carve-out for heartbeat may mask a legitimate fleet signal (operators silencing the fallback heartbeat). Either the rationale is wrong or step 6's 'not inherited' clause is wrong — they contradict each other.

Recommendation

Either (a) remove the heartbeat exclusion and trust the explicit-set semantics already defined in step 6, or (b) reword the constraint to give the real reason (e.g. 'heartbeat is the mandatory fallback — disabling it is operator-specific noise, not a default-flip signal') so the comment doesn't contradict the counting rule.

Counterpart finding

Notification link hard-codes main branch; breaks when default branch is not main

mediummaintainabilityhigh
  • skills/fork-skill-digest/SKILL.md:277
  • skills/fork-skill-digest/SKILL.md:280-281
Hard-coding main will 404 for repos that use a different default branch (e.g., master, trunk). This degrades notification UX and requires per-repo edits.

Recommendation

Use a branch-agnostic ref in the URL, e.g., /blob/HEAD/articles/fork-skill-digest-${today}.md, or detect the default branch via GITHUB_REF_NAME or gh api repos/$GITHUB_REPOSITORY and interpolate it.

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 →