AntFleet

Disagreement · 89395cd4-anthropic-4

History trim to 26 entries described in prose but not codified — risk of unbounded growth

solo Opus
repo 6f7fc663·PR #21·reviewed 1 week ago

Opus finding

History trim to 26 entries described in prose but not codified — risk of unbounded growth

lowdata-lossmedium
  • skills/contributor-spotlight/SKILL.md:217-219
  • skills/contributor-spotlight/SKILL.md:30-31
The 26-cap is described as English instructions only. The JSON example in Step 10 just shows an append, no trim command. An implementer who follows the example literally will append indefinitely. More importantly, the example fields don't match the prose: prose in Config says `{fork, featured_at, role}` but Step 10's example writes `{fork, owner, featured_at, cohort, enabled_count, stars_at_feature, operator_authored_count, commits_30d}` — eight fields, not three, and `role` is not present at all. Step 3 reads `contributor-spotlight-history.json` for dedup but doesn't specify which field it joins on (`fork`), so the schema drift creates a real risk that an early implementation writes `role` while later code reads `cohort` (or vice versa) and silently breaks 4-week dedup.

Recommendation

Pick one schema. Show a `jq` command that both appends and trims (`jq '.history = (.history + [$new] | .[-26:])' ...`) and reference the same field names in Step 3's dedup logic.

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.