Quick Start¶

This guide walks you through the most common claude-review workflows in under five minutes.

1. Set your API key¶

export ANTHROPIC_API_KEY=sk-ant-...

2. Review staged changes (default)¶

The simplest use case: stage your changes and run claude-review:

git add src/auth/token.go
claude-review

This reviews only the staged diff — exactly what git diff --cached would show.

Output is written to REVIEW.md in the current directory. Example:

# Code Review Report
Branch: feature/auth-refactor → main
Changes: +142 -38 across 8 files

## Summary
| Severity | Count |
|----------|-------|
| 🔴 Critical | 1 |
| 🟠 High | 2 |
| 🟡 Medium | 4 |
| 💡 Suggestion | 1 |

Cost: $0.43 · Agents: 5 finder + verifier + ranker · Time: 38s

---

## 🔴 Critical

### JWT expiry is never validated before granting access

**File**: `src/auth/token.go` · **Line**: 87 · **Confidence**: 95%

The decoded token's expiry (`exp` claim) is extracted but never compared against
the current time. Any expired token will be accepted as valid.

**Suggested Fix**:
if claims.ExpiresAt.Unix() < time.Now().Unix() {
    return nil, ErrTokenExpired
}

3. Review a specific commit or range¶

# Last commit
claude-review diff HEAD~1

# Last 3 commits
claude-review diff HEAD~3

# Branch diff against main
claude-review diff main..feature/auth-refactor

# Specific commit hash
claude-review diff a1b2c3d

4. Review a GitHub PR¶

# Full URL
claude-review pr https://github.com/org/repo/pull/123

# Just the PR number (auto-detects your remote)
claude-review pr 123

For PR reviews you need a GITHUB_TOKEN set:

export GITHUB_TOKEN=ghp_...

5. Review a GitLab MR¶

# Full URL
claude-review pr https://gitlab.com/org/project/-/merge_requests/45

# MR number shorthand (auto-detects GitLab remote)
claude-review pr 45

Set GITLAB_TOKEN before running:

export GITLAB_TOKEN=glpat-...

6. Review a Bitbucket PR¶

claude-review pr https://bitbucket.org/workspace/repo/pull-requests/7

Set BITBUCKET_TOKEN (an app password with PR read scope):

export BITBUCKET_TOKEN=...

7. Preview cost before running¶

Don't want to be surprised by the API cost? Use --estimate:

claude-review diff main..feature --estimate

This counts tokens and prints the estimated cost without making any API calls.

8. Output as JSON (for CI / bots)¶

claude-review diff HEAD~1 --format json --output review.json

The JSON output is structured for machine consumption — pipe it into Slack bots, dashboard scripts, or GitHub comment automation.

9. Focus on a specific concern¶

claude-review diff --focus security        # security issues only
claude-review diff --focus logic           # logic bugs only
claude-review diff --focus "security,types"  # multiple

10. Use a more powerful model for critical PRs¶

claude-review diff --model claude-sonnet-4-6    # deeper analysis
claude-review diff --model claude-opus-4-6      # maximum depth (most expensive)

11. Install as a pre-commit hook¶

Run reviews automatically before every commit:

claude-review install-hook

Remove it:

claude-review install-hook --remove

12. Enable memory across reviews¶

The memory layer remembers patterns across all your PRs. No setup required — just add --memory:

# Run a review with memory enabled
claude-review diff --memory
claude-review pr 123 --memory

After the first review, findings are stored in .claude-review/memory.db. On your next command, if 30+ minutes have passed or 10+ new findings exist, consolidation runs automatically in the background — no daemon needed.

# See what patterns have been detected across your PRs
claude-review insights

Learn more about the memory layer →