sourcebook

Catches the files your AI agent forgot to change.

INITIALIZE_ENVIRONMENT arrow_forward

$ npx sourcebook check

check_circle Modified: src/handlers/auth.ts, src/middleware/session.ts

⚠ HIGH: src/handlers/auth.test.ts

→ Test file for auth.ts was not updated

⚠ MEDIUM: src/config/permissions.ts

→ Co-change coupling 88% with auth.ts (12 shared commits)

2 FINDINGS Run with --ai for semantic analysis

0.4s — no LLM, pure structural analysis

Works with Claude Code · Cursor · GitHub Copilot · any AI coding agent

What is sourcebook?

AI agents don't just write wrong code. They write correct code and stop too early — leaving test files, sibling components, and co-dependent configs unchanged.

sourcebook checks your diff for completeness.
It flags the files your agent probably needs to change but didn't.

Under the hood, sourcebook maps your repo — import graphs, git co-change history, test file mappings, hub file detection. When you run sourcebook check, it compares your changes against that structural map and flags what's missing.

No LLM needed. Under a second. Runs as a CLI, a Claude Code hook, or an MCP server. For deeper semantic analysis, --ai sends the diff to Claude Sonnet for about a penny per run.

ALGORITHM_01

Import Graph PageRank

Ranks every file by structural importance. Find the hubs that break everything when you touch them.

Hub: types.ts (183 importers)
Hub: fixtures.ts (86 importers)
Hub: http.ts (72 importers)

ALGORITHM_02

Git Forensics

Reverted commits are "don't do this" signals. Co-change coupling reveals invisible dependencies.

CO-CHANGE: auth.ts ↔ session.ts
CORRELATION: 88%
REVERTS: 2 found (anti-patterns)

ALGORITHM_03

Convention Detection

Naming patterns, export style, barrel exports, path aliases — the tribal knowledge no README captures.

EXPORTS: named (75:18 ratio)
IMPORTS: path alias @/ preferred
BARREL: 40 index.ts files

FOUR_SURFACES

01_CLI

sourcebook check

Run it on any diff. Get told what's missing. Co-change coupling, test files, import graph siblings, hub blast radius. No LLM, under a second.

02_HOOKS

Claude Code hooks

sourcebook init wires up pre-commit hooks. Agent edits a file, sourcebook checks the diff, agent fixes it before committing.

03_MCP

MCP server

Published on the official MCP registry. Agents can query repo structure, blast radius, conventions, and co-change data on demand.

04_GITHUB_APP

PR completeness checks

Automated completeness review on every pull request. Comments with specific files that likely need to change. Coming soon for teams.

RESEARCH_FOUNDATION

ETH ZURICH, 2025

Auto-generated obvious context makes agents worse

LLM-generated context files reduced task success by 2-3% and increased inference costs by 20%+. Only non-discoverable information improves performance.

KARPATHY, 2026

program.md is the #1 lever for agent effectiveness

Autoresearch ran 700 experiments in 2 days because the curated context file contained only what the agent couldn't figure out alone.

AIDER

PageRank on import graphs for structural importance

Repo-map technique ranks files by how many other files depend on them. sourcebook uses this to identify architectural hubs.

CHROMA RESEARCH, 2026

LLMs lose 30%+ accuracy in the middle of long contexts

sourcebook places critical constraints at the top and bottom of output files — where LLMs pay the most attention.

SOURCEBOOK BENCHMARK, 2026

Agents that don't wander finish first

19 tasks, 10 repos, 4 languages. Repeat runs showed sourcebook eliminates the slow, wandering runs — and prevents failures handwritten context can't. Full methodology and results →

CHECK_YOUR_DIFF

Stop shipping incomplete changes. One command to find what's missing.

npx sourcebook check content_copy

VIEW_SOURCE

Free and open source. No LLM needed. BSL-1.1 licensed.