Second Opinion Code Review

The problem

Critical bugs like race conditions and data loss often survive standard reviews because the primary reviewer is too close to the code. Manual cross-model prompting is tedious, expensive if unoptimized, and often produces vague feedback without a structured plan.

What it does

• Assembles a minimized context pack by filtering out noise like lockfiles and build artifacts.
• Estimates token usage and API costs before execution to prevent runaway spending on large diffs.
• Executes a structured review request via external API with strict technical constraints and stack awareness.
• Categorizes findings by severity and provides concrete, line-specific fix proposals.

Frameworks & tools

Git, Python 3, curl, and jq. Compatible with Anthropic, OpenAI, or any LLM provider via API.

Why this beats prompting it yourself

It replaces "what do you think of this code?" with a repeatable, cost-aware audit pipeline. It enforces specific output formats and stack-based constraints that prevent generic AI advice and API-breaking suggestions.

Use cases

• Reviewing high-stakes paths like authentication, payments, or data migrations.
• Hunting for elusive edge cases in complex loops or concurrent operations.
• Getting a fresh perspective when a specific bug persists through manual QA.
• Validating a large refactor against existing public API constraints.

Known limitations

Requires an active LLM API key and command-line tools like jq. Not intended for style-checking or linting which should be handled by static analysis.