- Home
- Skills
- Technical Documentation
- Docs Drift Detector
Works with the AI Agents you already use
Docs Drift Detector
Reads one documentation file alongside the repo it describes and reports every claim that no longer matches the code
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Docs Drift Detector
Example session with this skill installed
check the README against the code before I share this repo with someone
- Read your context and instructions
- Compiled the docs drift detector
- Generated the document
Drift report - C:\Users\Nihil\Documents\GitHub\Stock-Image-Generator\README.md
Checked 40 claims - 21 stale, 1 unverified, 18 true
Stale
-
[install] "Together.ai API Key - Get one from Together.ai Dashboard" - README.md:22 (restated at :43, :77, :79)
Now: the app calls Pixazo's gateway with anOcp-Apim-Subscription-Keyheader; there is no Together.ai code path left anywhere insrc/-src/services/fluxApi.ts:68,src/services/fluxApi.ts:74,src/components/Configuration.tsx:240
Fix: "Pixazo API Key - Get one from the Pixazo Free API Dashboard" -
[install] "Node.js (v16 or higher)" - README.md:21
Now: Vite 5 requires^18.0.0 || >=20.0.0and sharp 0.33 requires^18.17.0 || ^20.3.0 || >=21.0.0; Node 16 cannot install or build this repo -package.json:26,package.json:35
Fix: "Node.js (v18.17 or higher)" -
[command] "This starts both the Electron main process and the React renderer in development mode." - README.md:63
Now: no script setsNODE_ENV, soisDevis false and Electron loads the prebuiltdist-renderer/index.htmlinstead of the Vite server it just started on :5173 -src/main.ts:19,src/main.ts:42,package.json:8
Fix: "This starts the Vite renderer on port 5173 and the Electron main process. Electron only attaches to the dev server whenNODE_ENV=developmentis set; otherwise it loads the last build fromdist-renderer/." -
[config] "Flux.1 Schnell Integration: Generate images using the state-of-the-art Flux.1 Schnell model via Together.ai" - README.md:7
Now: requests go tohttps://gateway.pixazo.ai/flux-1-schnell/v1/getData-src/services/fluxApi.ts:68
Fix: "Flux.1 Schnell Integration: Generate images using the Flux.1 Schnell model via the Pixazo Free API" -
[config] "Enter your Together.ai API key" - README.md:36
Now: the Configuration tab has a single key field labelled "Pixazo API Key" and expects a Pixazo subscription key -src/components/Configuration.tsx:229,src/components/Configuration.tsx:236
Fix: "Enter your Pixazo subscription key" -
[config] "Verify the API base URL (default:
https://api.together.ai)" - README.md:37 (restated at :80)
Now: the Configuration tab exposes no base-URL field at all; the endpoint is hardcoded in the service, and thebaseUrlstill sitting inDEFAULT_CONFIGis dead -src/components/Configuration.tsx:227,src/services/fluxApi.ts:68,src/App.tsx:19
Fix: "The API endpoint is fixed athttps://gateway.pixazo.ai/flux-1-schnell/v1/getDataand is not configurable." -
[config] "Confirm the model name (default:
black-forest-labs/FLUX.1-schnell)" - README.md:38 (restated at :81)
Now: no model field exists and no model is sent in the request body; the tab shows a read-only "Flux.1 Schnell (Free)" -src/components/Configuration.tsx:253,src/services/fluxApi.ts:48
Fix: "Confirm the model shown under Model Information - Flux.1 Schnell (Free). It is fixed and cannot be changed." -
[path] "│ └── upscaling.ts # Real-ESRGAN integration" - README.md:94
Now: the file is a Cloudinary upscaling client (CloudinaryUpscalingService,api.cloudinary.com) -src/services/upscaling.ts:19,src/main.ts:4
Fix: "│ └── upscaling.ts # Cloudinary upscaling integration" -
[path] "- Verify Real-ESRGAN executable is in the correct location" - README.md:111
Now: nothing insrc/references Real-ESRGAN; the installer script was gutted and says so -install-realesrgan.bat:1,src/services/upscaling.ts:19
Fix: "- Verify your Cloudinary cloud name, API key and API secret are set in the Configuration tab" -
[path] The
src/tree at README.md:85-101 presents itself as the layout
Now: it omitssrc/preload.js,src/hooks/,src/utils/,src/constants/, and three whole component subtrees (components/generation/,components/queue/,components/folders/) plus 8 top-level components; 4 of 22 components are listed -src/preload.js:1,src/hooks/useImageGeneration.ts:1,src/components/queue/PromptQueueManager.tsx:1
Fix: add the missing entries, or replace the tree with "Seesrc/-components/(UI, withgeneration/,queue/andfolders/subtrees),hooks/,services/,utils/,constants/,types/, plusmain.tsandpreload.jsfor the Electron side." -
[behaviour] " - 16:9 (Widescreen): 1344 x 768 px" - README.md:11
Now: 16:9 is 1024x576; the API rejects any side above 1024 -src/types/index.ts:102,src/services/fluxApi.ts:38
Fix: " - 16:9 (Widescreen): 1024 x 576 px" -
[behaviour] " - 3:2 (DSLR): 1152 x 768 px" - README.md:9, and the list of four at README.md:8
Now: 3:2 is 1024x680, and eight ratios ship - 2:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 1:1 - all rendered fromRESOLUTIONS-src/types/index.ts:87,src/types/index.ts:98,src/components/generation/GenerationControls.tsx:105
Fix: " - 3:2 (DSLR): 1024 x 680 px" (and extend the list to the eight ratios insrc/types/index.ts)
4 further stale claims not listed, all lower cost: the two-tab description (:17), the Vulkan GPU troubleshooting step (:112), "Verify you have sufficient API credits" (:117, the tier is free), and Real-ESRGAN in the licence list (:124, which also omits Cloudinary and Pixazo).
Unverified
- "npm run dist" packages the application - README.md:72 - needs: an actual electron-builder run. The script and config exist (
package.json:15), butelectron-builder.json:26points atbuild/icon.icoandbuild/contains onlyREADME.md, so packaging may fail on the missing icon.
Missing from the install path
- Build the renderer once before
npm run devornpm start-dist-renderer/is gitignored, so a fresh clone has nothing for Electron to load and opens a blank window -.gitignore:12,src/main.ts:42 - Add
build/icon.ico,build/icon.icnsandbuild/icon.pngbefore packaging -electron-builder.json:26,build/README.md:7 setup.bat/setup.shexist and do the install plus createtemp/andgenerated/, but Installation never mentions them -setup.bat:5,setup.sh:6
Verdict
No - a new user following this README will look for a Together.ai key the app cannot use, on a Node version that cannot build it, and land on a blank Electron window.
Sibling docs QUICKSTART.md, QUICK_INSTALL.md and BUILD.md repeat the same Together.ai and Real-ESRGAN claims; say the word and I will run them next.
Configuration.tsx
PDF · document
Example file from a real run - the skill writes it into your workspace.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Your README says one thing, the code does another. This skill reads a documentation file next to the repo it describes and returns every sentence that has stopped being true: the claim quoted with its line, what the code does now with the file and line that proves it, and the corrected sentence. Install steps are checked first, because a wrong one costs every new contributor an hour. Anything it cannot verify is listed as unverified, never guessed.
What it does
Reads one documentation file alongside the repo it describes and reports every claim that no longer matches the code - each one quoted with its line, the file and line that settles it, and the sentence that replaces it. Install and quickstart claims are checked first, because those are the ones that cost a stranger an hour.
Try it first
Open a repo whose README you suspect is a little behind and say "check the README against the code". You should get back a Drift report: the counts, a ranked stale list where every entry carries a doc line and a code line, an unverified list, anything missing from the install path, and a one-sentence verdict on whether a new user can follow the document today.
What it needs
An agent that can read and search the repo - Read, Grep, Glob and Bash. No API keys, no MCP servers, no network access. Language-agnostic, because it checks claims against files rather than parsing code.
It does not edit your documentation. It hands you the replacement sentence and stops, so you keep the final word on wording. It cannot settle a claim about anything outside the repo - a hosted service, a paid account, a machine you do not have - and lists those as unverified instead of guessing. Generated API references with hundreds of symbols are the wrong input; point it at the README, the setup guide or the contributing guide, where one wrong sentence actually breaks somebody's afternoon.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago