AntFleet

Disagreement · 1ea5c6c4-anthropic-0

code-reviewer PR listing filters in --json but selects by author after fetch; tail --jq path will include already-reviewed PRs

mismatch
repo 6f7fc663·PR #13·reviewed 1 week ago

Primary finding

code-reviewer PR listing filters in --json but selects by author after fetch; tail --jq path will include already-reviewed PRs

lowmaintainabilitymedium
  • templates/code-reviewer/SKILL.md:18-26
Step 1 says 'every open PR opened in the last 24h that hasn't been reviewed by this skill yet', but the gh pr list command does not filter by createdAt (no 24h window) and does not filter against the reviewed.json state file. The dedup against reviewed.json is described only in prose ('Skip anything already in there'). The 24h window mentioned in the comment is never actually applied — createdAt is fetched but never compared. This is a deceptive doc/comment vs. behavior: an operator copying this template gets a SKILL.md whose stated semantics (last 24h) are not implemented by the shown command.

Recommendation

Either add a --jq filter on createdAt (e.g. select((now - (.createdAt | fromdateiso8601)) < 86400)) or drop the '24h' wording. Also explicitly show the read of reviewed.json before the loop.

Counterpart finding

Fetches full PR diff before applying size gate (unnecessary work on large PRs)

lowperformancehigh
  • templates/code-reviewer/SKILL.md:29-34
The instructions fetch the full diff before checking additions+deletions against MAX_PR_LINES, causing unnecessary network and processing for large PRs that will be skipped anyway.

Recommendation

Fetch metadata first, evaluate additions+deletions, and only fetch the diff if under the threshold. Example: fetch meta; if ((additions+deletions <= MAX_PR_LINES)); then gh pr diff ...; fi.

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.

← back to all disagreementsview public receiptssee unanimous findings + anatomies →
Tweet ↗