// P06 — Code Review Agent // tree-sitter + Claude + static analysis · inline PR comments const P06_TERMINAL_SCRIPT = [ { t: 0.3, kind: 'caption', text: '01 / Clone the repository' }, { t: 0.8, kind: 'cmd', text: 'git clone git@github.com:Juadsuarezsan/code-review-agent.git', cwd: '~' }, { t: 2.6, kind: 'out', text: "Cloning into 'code-review-agent'..." }, { t: 2.95, kind: 'out', text: "remote: Enumerating objects: 134, done." }, { t: 3.25, kind: 'out', text: "Receiving objects: 100% (134/134), 228.8 KiB | 6.0 MiB/s, done." }, { t: 3.55, kind: 'out', text: "Resolving deltas: 100% (46/46), done." }, { t: 4.05, kind: 'cmd', text: 'cd code-review-agent', cwd: '~' }, { t: 5.4, kind: 'caption', text: '02 / Configure API keys' }, { t: 5.8, kind: 'cmd', text: 'cp .env.example .env' }, { t: 6.8, kind: 'cmd', text: '$EDITOR .env' }, { t: 7.6, kind: 'editor', title: '.env', lines: [ "ANTHROPIC_API_KEY=sk-ant-•••••••••••••••••", "ANTHROPIC_MODEL=claude-sonnet-4-5", "", "# GitHub (PR webhooks + posting comments)", "GITHUB_TOKEN=ghp_••••••••••", "", "# Storage + limits", "DATABASE_URL=postgresql://review:dev@localhost:5432/review", "MAX_COMMENTS_PER_PR=10" ] }, { t: 11.6, kind: 'out', text: '✓ .env saved · 0 secrets staged · API keys configured', color: 'success' }, { t: 13.0, kind: 'caption', text: '03 / Clone SWE-bench Lite (300 real GitHub issues)' }, { t: 13.4, kind: 'cmd', text: 'python scripts/download_swebench.py --variant lite' }, { t: 14.4, kind: 'out', text: '→ Pulling princeton-nlp/SWE-bench_Lite' }, { t: 15.0, kind: 'progress', start: 15.0, end: 18.0, total: 300, unit: 'issues', label: 'Loading' }, { t: 18.2, kind: 'out', text: '✓ 300 issues loaded (django, sympy, scikit-learn, requests, flask, ...)', color: 'success' }, { t: 18.5, kind: 'out', text: ' → for each: bug description · failing test · ground-truth fix patch', color: 'accent' }, { t: 19.2, kind: 'caption', text: '04 / Run review pipeline over all 300 issues' }, { t: 19.6, kind: 'cmd', text: 'python -m src.review.run_batch --workers 4' }, { t: 20.4, kind: 'out', text: 'INFO [batch] 5 parallel analyzers per PR: bug · style · tests · security · perf' }, { t: 20.8, kind: 'progress', start: 20.8, end: 34.4, total: 300, unit: 'PRs', label: 'Reviewing' }, { t: 34.6, kind: 'out', text: '✓ 300 reviews generated · 2,847 comments total · 9.5 avg per PR', color: 'success' }, { t: 35.0, kind: 'out', text: ' ├─ critical: 342 (12%)', color: 'danger' }, { t: 35.2, kind: 'out', text: ' ├─ high: 891 (31%)', color: 'warn' }, { t: 35.4, kind: 'out', text: ' ├─ medium: 1,168 (41%)', color: 'accent' }, { t: 35.6, kind: 'out', text: ' └─ low: 446 (16%)', color: 'success' }, { t: 36.4, kind: 'caption', text: '05 / Evaluate against SWE-bench ground-truth fixes' }, { t: 36.8, kind: 'cmd', text: 'python -m src.eval.bug_detection --top-k 3' }, { t: 38.0, kind: 'progress', start: 38.0, end: 44.0, total: 300, unit: 'PRs', label: 'Matching' }, { t: 44.2, kind: 'table', cols: ['system', 'bug recall', 'bug precision', 'F1', 'FP rate', '$/PR'], rows: [ ['ruff only', '0.18', '0.94', '0.30', '6%', '$0.00'], ['Claude single-pass', '0.61', '0.42', '0.50', '58%', '$0.04'], ['+ tree-sitter context', '0.73', '0.58', '0.65', '42%', '$0.07'], ['Pipeline (5 analyzers)', '0.79', '0.71', '0.75', '29%', '$0.12', 'highlight'], ]}, { t: 53.4, kind: 'caption', text: '06 / LLM-as-judge on 100 generated comments' }, { t: 53.8, kind: 'cmd', text: 'python -m src.eval.comment_quality --judge claude-opus-4' }, { t: 54.6, kind: 'out', text: 'INFO [eval] Rubric: actionable (1-5) · specific (1-5) · correct (1-5)' }, { t: 55.0, kind: 'progress', start: 55.0, end: 60.2, total: 300, unit: 'evals', label: 'Judging' }, { t: 60.4, kind: 'out', text: '✓ actionable 4.2 / 5', color: 'success' }, { t: 60.6, kind: 'out', text: '✓ specific 4.5 / 5', color: 'success' }, { t: 60.8, kind: 'out', text: '✓ correct 4.4 / 5', color: 'success' }, { t: 61.8, kind: 'caption', text: '07 / Publish GitHub Action to marketplace' }, { t: 62.2, kind: 'cmd', text: 'gh release create v1.0.0 --notes-from-tag' }, { t: 63.2, kind: 'out', text: '✓ Released v1.0.0 to Juadsuarezsan/code-review-agent', color: 'success' }, { t: 63.6, kind: 'out', text: '✓ Listed in GitHub Marketplace: "code-review-agent (Claude)"', color: 'accent' }, { t: 64.0, kind: 'cmd', text: 'cd frontend && pnpm dev' }, { t: 64.6, kind: 'out', text: ' ▲ Next.js 14.2.15 → ready in 1.1s', color: 'accent' }, { t: 64.9, kind: 'out', text: ' ✓ Local: http://localhost:3000 · 10 example PRs in gallery', color: 'success' }, { t: 65.6, kind: 'caption', text: '✓ Paste a diff or PR URL into the demo to see a review →' }, ]; const P06TerminalDemo = makeTerminalDemo({ script: P06_TERMINAL_SCRIPT, duration: 68, persistKey: 'p06-terminal', title: 'zsh — code-review-agent — 132×42', cwd: '~/code-review-agent', captionPrefix: 'P06', }); // ── System demo: diff viewer with inline comments materializing ──────── // Mock PR: a Django bug fix that has multiple issues. // 70s total. const P06_PR = { repo: 'django/django', number: 18241, title: 'Fix N+1 query in UserManager.get_users_with_pets()', author: 'jduarez', branch: 'fix-n-plus-one-users-with-pets → main', filename: 'django/contrib/auth/managers.py', // Diff lines: each with type (context, added, removed), line number, content diffLines: [ { t: 'context', oldN: 42, newN: 42, text: 'class UserManager(BaseUserManager):' }, { t: 'context', oldN: 43, newN: 43, text: '' }, { t: 'context', oldN: 44, newN: 44, text: ' def get_users_with_pets(self, *, active_only: bool = True):' }, { t: 'removed', oldN: 45, newN: null, text: ' users = self.filter(is_active=True) if active_only else self.all()' }, { t: 'removed', oldN: 46, newN: null, text: ' result = []' }, { t: 'removed', oldN: 47, newN: null, text: ' for user in users:' }, { t: 'removed', oldN: 48, newN: null, text: ' pets = user.pets.all() # N+1 query here' }, { t: 'removed', oldN: 49, newN: null, text: ' if len(pets) > 0:' }, { t: 'removed', oldN: 50, newN: null, text: ' result.append((user, pets))' }, { t: 'removed', oldN: 51, newN: null, text: ' return result' }, { t: 'added', oldN: null, newN: 45, text: ' qs = self.filter(is_active=True) if active_only else self.all()' }, { t: 'added', oldN: null, newN: 46, text: ' qs = qs.prefetch_related("pets").filter(pets__isnull=False)' }, { t: 'added', oldN: null, newN: 47, text: ' return list(qs.distinct())' }, { t: 'context', oldN: 52, newN: 48, text: '' }, { t: 'context', oldN: 53, newN: 49, text: ' def deactivate_inactive_users(self, days: int):' }, { t: 'added', oldN: null, newN: 50, text: ' cutoff = timezone.now() - timedelta(days=days)' }, { t: 'added', oldN: null, newN: 51, text: ' return self.filter(last_login__lt=cutoff).update(is_active=False)' }, ], // Inline AI comments appearing at specific line numbers comments: [ { t: 1.8, line: 46, severity: 'high', analyzer: 'bug', title: 'Returns Users without pets', body: '`pets__isnull=False` excludes users with no pets, but the original code returned `(user, pets)` tuples — the return type is now `list[User]`, not `list[tuple[User, QuerySet[Pet]]]`. Callers will break.', suggestion: 'Either preserve the tuple shape, or update the docstring + type hint and bump the major version of this internal API.', }, { t: 7.0, line: 47, severity: 'medium', analyzer: 'perf', title: '`distinct()` may not be needed', body: '`prefetch_related` doesn\'t produce duplicates, but `filter(pets__isnull=False)` joins, which can. Confirm with EXPLAIN: if the query already returns unique rows, `distinct()` adds an unnecessary `GROUP BY`.', }, { t: 13.0, line: 50, severity: 'high', analyzer: 'bug', title: 'Missing import: `timezone`, `timedelta`', body: '`timezone.now()` requires `from django.utils import timezone`; `timedelta` requires `from datetime import timedelta`. Neither is imported in this file — will fail at runtime on the first call.', }, { t: 20.0, line: 51, severity: 'critical', analyzer: 'security', title: 'Mass deactivation without audit trail', body: 'A single call to `deactivate_inactive_users(0)` would disable every user instantly. Add a hard floor (e.g. raise on `days < 7`), log the affected count, and write an audit row per change.', }, { t: 28.0, line: 49, severity: 'low', analyzer: 'tests', title: 'No test coverage for new method', body: 'New public method on the manager with no corresponding test in `tests/auth_tests/test_managers.py`. CI coverage check will pass (file already > 70%), but the method itself has 0% coverage.', suggestion: 'Add `test_deactivate_inactive_users_respects_days_floor` covering both the success path and the floor enforcement.', }, { t: 36.0, line: 47, severity: 'low', analyzer: 'style', title: 'Inconsistent return type with sibling methods', body: 'Other `UserManager` methods return `QuerySet`, not `list`. Returning `list(...)` here forces eager evaluation — callers can no longer chain `.filter()` or `.order_by()`. Consider returning the queryset.', }, ], summary: { t: 44, total: 6, critical: 1, high: 2, medium: 1, low: 2, verdict: 'CHANGES REQUESTED', score: 0.62, }, }; const ANALYZERS = [ { id: 'bug', label: 'Bug Detector', icon: '🐛', color: '#ff5a72' }, { id: 'style', label: 'Style Checker', icon: '✎', color: '#9aa3b8' }, { id: 'tests', label: 'Test Gap Analyzer', icon: '✓', color: '#22d3ee' }, { id: 'security', label: 'Security Scanner', icon: '⚿', color: '#fbbf24' }, { id: 'perf', label: 'Perf Reviewer', icon: '⚡', color: '#f472b6' }, ]; function P06SystemDemo() { return (
); } function P06TopBar() { const t = useTime(); const visibleComments = P06_PR.comments.filter(c => t >= c.t); return (
{P06_PR.repo} #{P06_PR.number} · {P06_PR.title}
@{P06_PR.author} · {visibleComments.length} comments
); } function P06DiffPanel() { return (
{P06_PR.filename} {P06_PR.branch}
{P06_PR.diffLines.map((l, i) => )}
); } function P06DiffLine({ line, index }) { const t = useTime(); // Find comment whose line === this newN (or oldN for removed lines) const comment = P06_PR.comments.find(c => c.line === line.newN); const showComment = comment && t >= comment.t; const bg = line.t === 'added' ? 'rgba(52,211,153,0.08)' : line.t === 'removed' ? 'rgba(255,90,114,0.10)' : 'transparent'; const mark = line.t === 'added' ? '+' : line.t === 'removed' ? '-' : ' '; const markColor = line.t === 'added' ? '#34d399' : line.t === 'removed' ? '#ff5a72' : '#5a5f6e'; return ( <>
{line.oldN ?? ''} {line.newN ?? ''} {mark} {line.text}
{showComment && } ); } function P06InlineComment({ comment }) { const sevColor = comment.severity === 'critical' ? '#ff5a72' : comment.severity === 'high' ? '#fbbf24' : comment.severity === 'medium' ? '#22d3ee' : '#9aa3b8'; const analyzer = ANALYZERS.find(a => a.id === comment.analyzer); return (
{comment.severity.toUpperCase()} {analyzer.icon} {analyzer.label} Claude · review-agent
{comment.title}
{comment.body}
{comment.suggestion && (
SUGGEST: {comment.suggestion}
)}
); } function P06SidePanel() { const t = useTime(); // Analyzer activations — each becomes "done" once any of its comments appears return (
parallel analyzers
{ANALYZERS.map(a => { const acComments = P06_PR.comments.filter(c => c.analyzer === a.id); const visibleCount = acComments.filter(c => t >= c.t).length; const earliest = acComments.length ? Math.min(...acComments.map(c => c.t)) : 999; const active = t >= 1 && t < earliest && acComments.length > 0; const done = visibleCount === acComments.length && acComments.length > 0; return (
{a.icon}
{a.label}
{active ? 'scanning...' : done ? `${visibleCount} comment${visibleCount===1?'':'s'}` : visibleCount > 0 ? `${visibleCount}/${acComments.length}` : '—'}
{done && ✓}
); })}
{t >= P06_PR.summary.t && (
→ {P06_PR.summary.verdict}
)}
); } function SummaryStat({ n, l, c }) { return (
{n}
{l}
); } Object.assign(window, { P06TerminalDemo, P06SystemDemo });