awesome-quant/.claude/skills/bprr/SKILL.md

name, description

name	description
bprr	Bulk PR reviewer for awesome-quant. Use when the user asks to review all open PRs, review unreviewed PRs, bulk review, or mentions "bprr". Reviews open PRs lacking the reviewed label and presents a summary before any merge/comment/label action.

BPRR: Bulk PR Reviewer

Review multiple open pull requests for README entry contributions.

Hard Rules

1. Use GitHub MCP tools for PR operations. Do not use gh for PR review, comments, labels, closing, or merging.
2. Filter out PRs with the reviewed label unless the user asks to re-review them.
3. Do not comment, label, close, or merge until the user explicitly approves each action.
4. Do not auto-merge all approved PRs unless the user explicitly selects that option.
5. If GitHub MCP tools are unavailable, report the blocker and point to docs/codex-setup.md.

Workflow

1. List open PRs sorted oldest first.
2. Filter to unreviewed PRs by default.
3. Fetch the default-branch README.md once for duplicate checks.
4. For each PR, fetch details, current head SHA, labels, files, body, diff, and the latest Validate PR workflow/check attempt for that SHA.
5. Review only added README.md entries unless the PR changes other files; flag other file changes as unusual.
6. Apply the CI evidence rules below, then perform the remaining $sprr checks. Use scripts/validate_readme.py --diff-from <base-ref> only when current-head CI does not establish mechanical validity and the PR branch is available locally; otherwise apply the validator's rules manually from the MCP diff.

CI Evidence Rules

Evaluate each PR independently. Use only the latest attempt for the current head SHA; a newer queued or pending attempt supersedes an older success for that SHA. A result for one PR or commit never applies to another.

Current-head result	Review behavior
success	Accept parser format, tag syntax, separators, the required final period, HTTPS, GitHub-link syntax, recognized section, and base-README duplicate checks as passed. Do not repeat those mechanical checks.
failure	Inspect the failing job or step. Use NEEDS CHANGES when validation failed; if another step failed or details are unavailable, reproduce mechanical validation before deciding.
queued, pending, awaiting approval	Mark the review incomplete and do not return APPROVE.
skipped, cancelled, missing	Treat mechanical validation as unverified and reproduce it before deciding.
success only on an older SHA	Report STALE, ignore it, and reproduce mechanical validation for the current head SHA.

For every PR, still inspect the diff and manually review tag meaning and concision, description quality, relevance, semantic section suitability, commercial classification, repository activity/archive/documentation/community evidence, duplicates in open PRs and PRs closed within the last 365 days, and multi-project relatedness.

Validation Checklist

For each added entry, check:

• Parser regex match.
• Required backtick tags for new non-commercial entries. Treat them as a compact tag cloud: accept concise languages, runtimes, protocols, interfaces, data types, and domain terms.
• Separate concepts must use adjacent tags, such as `Python` `C++` `MCP`; do not require every tag to be a programming language.
• Description period before optional [GitHub](...).
• https:// URLs.
• Exact optional [GitHub](https://github.com/owner/repo) format.
• Correct category section.
• Commercial placement under Commercial & Proprietary Services.
• Duplicate project names or URLs.
• Treat any verifiable GitHub repository mentioned as the main URL or exact [GitHub](...) suffix as a strong positive relevance signal.
• For GitHub repos, check source availability, activity, archived status, documentation, and community evidence. GitHub relevance does not waive duplicate, format, or quality checks.
• Clear rationale for multiple related projects in one PR.

Summary Output

Present a table:

| PR | Title | Author | Head | Validate PR | Entries | Format/URL | Section | Duplicate | Verdict |
|----|-------|--------|------|-------------|---------|------------|---------|-----------|---------|

Use the entry count in Entries. Use PASS, FAIL, PENDING, UNVERIFIED, or STALE in Validate PR. Use CI PASS, CI FAIL, REPRODUCED PASS, REPRODUCED FAIL, or INCOMPLETE in Format/URL. Section remains a manual semantic judgment. In Duplicate, report both README PASS|FAIL and PR SEARCH PASS|FAIL.

After the table, list any PRs that need detailed notes, including the failing workflow step or the reason CI evidence was not accepted.

Ask what to do next. Accept selections like:

• merge 123, 124
• comment 125
• close 126
• all approved
• none

Confirm each merge before executing it.