From vgv-wingspan
Runs parallel quality review agents on files, directories, branch diffs, or full projects to assess code quality and identify issues before merging.
npx claudepluginhub verygoodopensource/very_good_claude_code_marketplace --plugin vgv-wingspanThis skill is limited to using the following tools:
Run quality review agents. Review manually written code, assess existing codebases, or check a branch before merging.
Orchestrates parallel multi-agent code reviews with ≥80% confidence filtering for quality, security, and auto-detected discipline-specific issues via git diffs.
Orchestrates parallel agents to review code changes for CLAUDE.md compliance, bugs, git history, and security with confidence scores. Outputs reports or posts to GitHub PRs.
Reviews code changes using parallel personas for correctness, testing, maintainability, and conditional areas like security, performance, APIs. Merges into P0-P3 severity reports for PR prep and iterative feedback.
Share bugs, ideas, or general feedback.
Run quality review agents. Review manually written code, assess existing codebases, or check a branch before merging.
<review_scope>$ARGUMENTS</review_scope>
Parse the review scope above for optional file paths or directories.
If paths are provided:
If no paths provided:
Run the scope detection script:
${CLAUDE_SKILL_DIR}/scripts/detect-review-scope.sh
SCOPE=branch: use the listed files as review scope. Announce scope summary: number of changed files, which areas of the codebase are affected. Proceed to Step 2.SCOPE=default: tell the user: "You're on <CURRENT_BRANCH>. No branch diff available." Use AskUserQuestion: "What would you like to review?" with options:
Run pwd and let <PWD> be the result — subagents may change directories, making relative paths unreliable. The report directory is <PWD>/docs/code-review/.
Run the default review agents listed below in parallel. Projects may define additional review agents in their CLAUDE.md — if any are specified, include them alongside the defaults. Projects may also replace the default set entirely by specifying their own list.
Each agent prompt must include:
The scope constraint — one of:
The report output instructions (substitute <PWD> with the absolute path resolved above):
Write your full detailed report to
<PWD>/docs/code-review/<name>.md(create the directory if needed). This is an absolute path — use it exactly as given, do not convert to relative. Then return ONLY a short structured summary to the parent context in this format:## <Agent Name> Summary **Report**: `<PWD>/docs/code-review/<name>.md` (<word_count> words) **Critical**: <count> | **Important**: <count> | **Suggestions**: <count> ### Findings - [Critical] <one-line description> - [Important] <one-line description> - [Suggestion] <one-line description>Do NOT return the full report text. Only return the summary above.
Default agents and their report filenames (substitute <PWD> with the absolute path resolved above):
| Agent | Report file |
|---|---|
| @vgv-review-agent | <PWD>/docs/code-review/vgv-review.md |
| @code-simplicity-review-agent | <PWD>/docs/code-review/code-simplicity-review.md |
| @test-quality-review-agent | <PWD>/docs/code-review/test-quality-review.md |
| @architecture-review-agent | <PWD>/docs/code-review/architecture-review.md |
If an agent fails: Note the failure, continue with successful agents. After all agents complete, report which (if any) failed and offer to retry.
After all reviews complete:
Categorize findings from all summaries into Critical, Important, and Suggestions.
Present the consolidated summary to the user with counts per category and the one-line descriptions from each agent.
If no findings: Code looks good. Reports are at docs/code-review/.
Use AskUserQuestion to present post-review options:
docs/code-review/ for manual review. Done.After fixing (if chosen):
docs/code-review/ already exists from a previous review, old reports will be overwritten by agents with the same name. Delete the directory first if you want a clean slate.docs/code-review/ as untracked working files. Commit or delete them when no longer needed.