Run an explicitly requested Zephyr control/treatment benchmark on the same pre-normalized sample and compare Finelog stage metrics.
日本語の概要は準備中です。原文の説明を表示しています。
Review an explicitly identified Marin pull request against the repository's correctness and maintainability expectations.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Review the requested pull request at the exact head named by the caller. The
review is read-only except for comments posted when --comment is requested.
A reviewable Marin pull request:
AGENTS.md or CLAUDE.md instructions. If
tests change, it also follows root TESTING.md and any scoped testing guide..agents/skills/writing-style/pull-requests.md; the body is
the squash-merge commit message, not a template, file inventory, or test log.Report only actionable defects introduced by the pull request. Validate every
finding against the diff, relevant surrounding code, and applicable
instructions. Do not report style preferences, speculative failures, harmless
duplication in experiments/grug, or pre-existing problems. If the evidence is
uncertain, omit the finding.
Inspect the pull request metadata, issue comments, inline comments, and current head. Stop without commenting when the pull request is closed, a draft, an automated dependency update, or too trivial to benefit from code review. Still review agent-generated pull requests.
Check both issue and inline comments for
<!-- marin-correctness-review -->. Stop when a prior automated correctness
review is present unless a maintainer explicitly requested another review.
Confirm that the checked-out commit and the pull request's current head both match the requested head. If either differs, report the stale review and post nothing.
Read the root instructions and every instruction file that scopes a changed path. Read the root and scoped testing guides when tests change. Inspect the complete diff and enough surrounding code to validate behavior and call-site consistency.
Review the code against the standard above. For each finding, record its
changed file and line, category (bug, instruction-following, or
maintainability), concrete impact, and the evidence that makes it certain.
Separately record title or description problems.
Emit one best-effort stats event before returning, including clean and non-commenting reviews. Never retry or surface a telemetry failure. Run from the repository root:
cat <<'EOF' | uv run infra/codehealth/log_stats.py
{
"tool": "review-pr",
"invocation": {
"trigger": "local",
"agent_cli": "codex",
"pr_number": <PR>,
"agent_exit_code": 0,
"timed_out": false
},
"findings": [
["<file>", <line>, "<category>", 1.0, "<first 200 chars of issue description>"]
]
}
EOF
Use an empty findings array when the code review is clean.
Print the findings or No issues found. and list pull-request-description
problems separately. Without --comment, stop here.
With --comment, post one top-level comment for all title or description
problems. Begin with 🤖, name each concrete problem and fix, and end with
<!-- marin-correctness-review -->.
If the code review is clean and step 8 did not post a description comment, post exactly this top-level comment. Otherwise stop after the description comment.
🤖 Code review
No issues found.
<!-- marin-correctness-review -->
Otherwise, post one inline comment per unique code finding. Begin each body
with 🤖, explain the defect and its impact, and end with
<!-- marin-correctness-review -->. Include a committable suggestion only
when that suggestion completely fixes a small, self-contained issue. Use a
full-SHA GitHub link with surrounding context when citing repository
instructions or related code.
Use gh for GitHub reads and top-level comments. Post inline comments with the
GitHub inline-comment tool and confirmed: true.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Run an explicitly requested Zephyr control/treatment benchmark on the same pre-normalized sample and compare Finelog stage metrics.
日本語の概要は準備中です。原文の説明を表示しています。
Register a named Hugging Face dataset for Marin by inspecting its schema and adding the appropriate experiments/datasets module.
日本語の概要は準備中です。原文の説明を表示しています。
Implement a fully specified request for a GCP IAM resource grant or Cloud Run IAP viewer in marin-iac.
日本語の概要は準備中です。原文の説明を表示しています。
Add or change a named Pallas/Mosaic kernel, including its reference implementation, correctness tests, wrapper, or requested tuning.
日本語の概要は準備中です。原文の説明を表示しています。
Archive explicitly named legacy experiment files by publishing a dated tag and canonical issue breadcrumbs.
日本語の概要は準備中です。原文の説明を表示しています。
Launch or continuously monitor a specified Zephyr pipeline on Iris only when asked to babysit it; restart only after explicit approval.
日本語の概要は準備中です。原文の説明を表示しています。