Docs that stay true to your code.

Staleproof follows every pull request through a graph of your code and docs to the paragraphs it makes false, runs the examples to prove it, and writes the fix.

It also checks the files AI coding agents read: a stale command, link or env var in AGENTS.md, CLAUDE.md or your Cursor and Copilot rules is flagged like any other doc. How lint works

Add to GitHub Check a pull request

$npx -y staleproof check https://github.com/FalkorDB/falkordb-py/pull/262

Open source, MIT
No model key needed
Python · TypeScript · JavaScript · Go

FalkorDB/falkordb-py#262fix: only forward accepted kwargs from Is_Cluster

changedIs_Cluster
called byFalkorDB.__init__
used inREADME · Asynchronous Example

beforefailsTypeError: unexpected keyword argument 'himport_registry'

afterpassesexit 0

No doc names Is_Cluster, so a search finds nothing. The call graph finds the README.

Measured, not promised

Every false positive is counted in the open on the Drift Hall

33 / 34

findings confirmed real on 18 repositories Staleproof had never seen

Drift Hall · held-out

41 of 43

doc sections one pull request falsified were in other repositories

falkordb-py#276

0 / 47

of the sections the graph reaches would be found by a grep for the changed names

falkordb-py #262 · #267 · #276

~1 ms

to answer "which paragraphs can this change make false?" on FalkorDB

impact query

Try it

Check any pull request

Paste a pull request from any public Python, TypeScript or JavaScript repository. The first check indexes it in seconds.

Graph mode

Impact, clearance, ranking and example runs in a fixed order. No model, no key, deterministic.

Agent mode

A model reads each affected section and writes a verdict or an edit for every one.

The interactive demo needs JavaScript. These are the pull requests it opens with:

FalkorDB/falkordb-py#262 — fix: only forward accepted kwargs from Is_Cluster to redis.Redis
FalkorDB/falkordb-py#267 — test: assert the client parses execution plans, and fix a plan parser bug
FalkorDB/falkordb-py#276 — fix(asyncio): prevent connection pool kwargs mutation and await schema refresh

Who it's for

Anyone whose docs live apart from their code

Library and SDK maintainers

Your client, the SDK built on it and the docs site are three repositories. A change in the first surfaces the pages in the other two.

41 of 43 sections falkordb-py#276 reached were in other repositories.

GitHub App

Teams coding with AI agents

Claude Code, Cursor and Copilot read AGENTS.md and CLAUDE.md. Staleproof keeps those files true, and answers your agent over MCP before it changes a function.

docs_for_symbol, check_changes and the docs patch, in the same session as the code.

MCP server

DevRel and docs teams

Know what a release falsified, most-read page first, and ship one patch with the evidence behind every edit.

An audit lists the public API no page covers, and every example that no longer fits the code.

Audit a repository

Open-source maintainers

A docs check on every pull request: the sections it can make false, the path for each, and a button that applies the fix.

Beta rules only inform. Nothing fails a check until the Drift Hall shows it at 80% precision or better.

GitHub Action

Why it works

Three things a search can't do

It follows calls, not names

Up to three call hops from the changed function, across repositories, to the docs that use it. Grep stops at the name.

It runs the example, before and after

Each affected example runs at both commits in a sandbox. "This PR broke it" is a test result, not a guess.

Every finding carries its receipt

The path from the change to the paragraph is shown on every hit. No path, no edit: the tool refuses.

How it works

From a pull request to a proven docs fix

01
Index
Code, docs and examples from every repository go into one FalkorDB graph, with the calls between them.
02
Trace
From each changed function, walk the callers to the sections that use them. The rest is cleared, with proof.
03
Prove
Run each affected example at the commit before the PR and at the PR, in a sandbox.
04
Fix
Every section gets a verdict, and a git apply-able patch with the evidence. Decisions are remembered for the next PR.

The question is a graph question

"Which paragraphs, in which repositories, can this change make false?" is reachability over call edges. One Cypher pattern answers it.

The graph model and every query

impact (excerpt)

MATCH (pr:PullRequest {url: $pr})<-[c:CHANGED_IN]-(s:Symbol)
WHERE c.kind <> 'added'
MATCH p = (s)<-[:CALLS*0..3]-(t:Symbol)<-[:MENTIONS|USES]-(d)
WHERE (d:DocSection OR d:CodeExample) AND d.status = 'live'
RETURN d.id, length(p) AS hops,
       [n IN nodes(p) | coalesce(n.qualname, n.id)] AS evidence

Install

Six ways in, one npm package

Pick the one that matches where your docs break.

Install once, every pull request is checked

No workflow file and no secrets. Each PR gets a check with annotations, suggestions, and an Apply fixes button.

GitHub App guide

github.com/apps/staleproof

1. Install the app on your repositories
2. Open a pull request
3. Read the "Staleproof" check: the docs it can make false,
   the path for each, and the fix as a suggestion

One step, no inputs

The job starts its own FalkorDB, indexes the repository, runs the affected examples before and after, and posts one comment.

GitHub Action guide

.github/workflows/staleproof.yml

on:
  pull_request:
permissions:
  contents: read
  pull-requests: write
jobs:
  docs:
    runs-on: ubuntu-latest
    steps:
      - uses: rajkaria/staleproof@v1

From nothing to a checked pull request

No install and no Docker: the first check starts FalkorDB locally and indexes the repository. lint needs no graph at all.

CLI guide

bash

$ npx -y staleproof lint            # paths, links, commands, env, symbols
$ npx -y staleproof check https://github.com/FalkorDB/falkordb-py/pull/262
$ npx -y staleproof fix             # applies only verified fixes

Ask Claude Code or Cursor before you change a function

The hosted server needs nothing installed and indexes any public repository on first use. Every result lists the Cypher it ran.

MCP guide

bash

$ claude mcp add --transport http staleproof https://staleproof.dev/api/mcp

> Which docs depend on Is_Cluster? Which examples do my edits break?

Diagnostics as you type

Findings from unsaved text, fixes as code actions, hover and go-to-definition from a doc to the code it names. Cursor, Windsurf, Zed, Neovim and JetBrains too.

VS Code guide

bash

$ code --install-extension staleproof.staleproof

Stream a check over server-sent events

serve runs this web app and its JSON API; the package also exports the indexer, the tools and both check modes as a library.

API guide

bash

$ npx -y staleproof serve --port 8080
$ curl -N localhost:8080/api/check \
    -H 'content-type: application/json' \
    -d '{"pr_url": "https://github.com/FalkorDB/falkordb-py/pull/262", "mode": "graph"}'

Compared

Why not grep, or a vector store?

	grep / code search	Vector search	Staleproof
Finds docs that reach the change through other functions	Only the names you search for	Similar text, not reachability	Up to 3 call hops, any repository
Proves a change broke an example	No	No	Runs it before and after
Says why a paragraph is flagged	The matching line	A similarity score	The evidence path
Noise on 4 checked pull requests	227 unrelated matches	—	Only sections the change reaches

FAQ

Questions developers ask first

Do I need an LLM API key?

No. Graph mode runs impact, clearance, ranking and example runs with no model. Agent mode adds a model that reads each section and writes the verdicts and edits; it needs ANTHROPIC_API_KEY or any Anthropic-compatible endpoint.

Which languages does it understand?

Code: Python, TypeScript, JavaScript and Go, across repositories. Docs: Markdown, MDX and reStructuredText. Python examples are run before and after a pull request; JavaScript and TypeScript examples are checked statically.

Do I need a pull request?

No. Audit any public repository to see which examples no longer fit the code today, run lint in a pre-commit hook, or ask over MCP which docs depend on a function before you change it.

Does my code leave my machine?

With the CLI, the Action or the editor extension, indexing, the graph and example runs stay where you run them; private repositories work with a token. Graph mode sends nothing to a model. The hosted site and MCP server index public repositories only.

Will it open pull requests on its own?

No. It proposes edits as suggestions, a git apply-able patch, or one fix PR you opt into. Nothing is pushed without you.

What does it cost?

Nothing: it is MIT licensed. You run FalkorDB (locally, in Docker, or on FalkorDB Cloud's free tier) and pay your model provider only in agent mode.

Check your next pull request before your readers do.

Install once. Every pull request after that gets its docs impact, with proof.

Add to GitHub Star on GitHub

Docs that stay true to your code.

Measured, not promised

Check any pull request

Anyone whose docs live apart from their code

Library and SDK maintainers

Teams coding with AI agents

DevRel and docs teams

Open-source maintainers

Three things a search can't do

It follows calls, not names

It runs the example, before and after

Every finding carries its receipt

From a pull request to a proven docs fix

Index

Trace

Prove

Fix

The question is a graph question

Six ways in, one npm package

Install once, every pull request is checked

One step, no inputs

From nothing to a checked pull request

Ask Claude Code or Cursor before you change a function

Diagnostics as you type

Stream a check over server-sent events

Why not grep, or a vector store?

Questions developers ask first

Check your next pull request before your readers do.