DecisionLint
Lint Jev and Luna questions before they are sent: the questions you write for TypeSafe's Jev, OpenAI's Decisions API (gpt-6-luna) or any platform that serves one, as you type, with no API key and no network calls. An optional command asks the model itself to check a file
No editor in the loop? npx -y decisionlint-mcp runs the same engine as an MCP server, so an agent can call lint_text with nothing else installed. Every way to install it.
Overview
What DecisionLint does
Jev and Luna answer the question you wrote, which is not always the one you meant. DecisionLint finds the questions a project sends to a decision model, TypeSafe's Jev, OpenAI's Decisions API or any platform that serves either shape, from OpenRouter and Vercel AI Gateway to a llama.cpp on your own machine, in JSON, JavaScript, TypeScript, Python, Rust and Go, and reports the ones written in a way measured to produce bad answers: a choice with no fallback option, a score given a map where it takes a list, a type that does not exist, a question that turns on a word with no stated line. The limits a request is held to are the platform's own, read from the model id, a URL in the file or a target setting. Linting runs as you type with no API key and no network. A part of a question that is built at runtime is counted as not read, never guessed at. Two optional commands, run by you with your own key at whichever platform you send to, ask the model itself whether a file's questions have problems a text pattern cannot see, or re-send one question with its layout changed to see whether the answer holds.
What people use it for
Catch it while typing
Findings appear in the editor as a question is written, with a quick fix where the fix is not a guess.
Check a pull request
The same checks run from the command line, with output for a terminal, as JSON, or as GitHub annotations.
Hand it to an agent
The MCP server offers lint_text, lint_paths and list_rules, so an agent can check the questions it writes.
Palette
19 commands
Open the command palette and type the name. Nothing is bound to a shortcut by default — the editor's keymap is the user's, not ours.
- Lint the Questions in This File
decisionlint.lintFile - Lint the Questions in the Workspace
decisionlint.lintWorkspace - Open Settings
decisionlint.openSettings - Ask Jev About This File (TypeSafe, uses your API key)
decisionlint.askJev - Ask Luna About This File (OpenAI, uses your API key)
decisionlint.askLuna - Ask via OpenRouter About This File (uses your API key)
decisionlint.askOpenRouter - Ask via Vercel AI Gateway About This File (uses your API key)
decisionlint.askVercel - Ask via LLM Gateway About This File (uses your API key)
decisionlint.askLlmGateway - Ask Liquid d1 About This File (uses your API key)
decisionlint.askLiquid - Ask the Model at My Endpoint About This File
decisionlint.askEndpoint - Set TypeSafe API Key
decisionlint.setTypeSafeApiKey - Set OpenAI API Key
decisionlint.setOpenAIApiKey - Set OpenRouter API Key
decisionlint.setOpenRouterApiKey - Set Vercel AI Gateway API Key
decisionlint.setVercelApiKey - Set LLM Gateway API Key
decisionlint.setLlmGatewayApiKey - Set Liquid AI API Key
decisionlint.setLiquidApiKey - Set Endpoint API Key
decisionlint.setEndpointApiKey - Clear API Keys
decisionlint.clearApiKeys - Probe the Question at the Cursor (asks Jev, uses your TypeSafe key)
decisionlint.probeQuestion
Rules
37 rules
Each code links to its page: what it catches, an example, and how to fix or silence it.
Requests Read from the text, offline.
Part of the question is built at runtime, so the rules that need it did not run
An alias such as
jev-latestmoves with each release, so answers can change with no change on your sideA Choice has more than 255 options. The API rejects it
A Score has more than 10 levels. The API rejects it
A Choice has no
otherornone of the above, so an input that fits no option is forced into one. The largest cost measured anywhere, and the hardest to tell from textA question id, option or level appears twice. In an object the later one silently replaces the earlier
A Choice needs a map of options, a Score needs an array of levels, and a Noul takes
trueandfalseThe type is missing or unknown, or the instructions are an empty string
Score levels are bare numbers. Jev matches the state against each description and never sees its position
A Choice with one option or a Score with one level gives every input the same answer
A
decisionlint-disablecomment silences nothing, so it can only hide a finding added laterAn option's description only repeats its name, or a Noul's criteria say yes and no. Jev matches the state against the description, and this one says nothing
The instructions are one or two words, so the question leans on its id. Jev is sent the instructions and never the id
modelnames a Jev version newer than the one these rules were measured on, so a finding may cost nothing there and a weakness new to it is not looked forA request has more questions than the platform it goes to takes in one call. The platform rejects it
Wording Read from the text, offline.
A question negates twice in one clause, or a negation sits directly on another
The question asks Jev to count or compare numbers
The question asks Jev to order two times or measure the gap between them
A Noul joins two judgments with 'and'
The question asks for a value or for text to be written
A Noul with no criteria is phrased so that yes means something is absent
A Noul's 'true' criterion describes the negative case
Score levels are degree words, bare labels, or one word turned up and down. Nothing for Jev to match the state against
The question refers to a value by hex or RGB encoding
A Noul with no criteria turns on a word such as large, often or enough
Asked of the model Run only when you ask, with your own key.
Jev reads the question as needing counting or arithmetic
Jev reads a Noul as turning on a matter of degree with no stated line
Jev reads two options of a Choice as covering the same cases
Jev reads two levels of a Score as the same situation
Jev reads an option as named for one thing and described as another
Jev reads the criteria as deciding something the instructions do not ask
Jev reads a Choice's options as steps on one scale, which a Score would place between
Jev reads a Score's levels as unordered categories, which a Choice would pick from
Jev reads a question as needing another question's answer from the same request
Jev reads two questions in one request as asking for the same judgment
Jev reads the state written in the file as not holding what the question asks about
Jev reads part of the state written in the file as giving orders to its reader
Install
Install DecisionLint
VS Code
ext install nolindnaidoo.decisionlintOpen Quick Open (Cmd/Ctrl+P) and paste.
Cursor / Windsurf
cursor --install-extension nolindnaidoo.decisionlintCursor, Windsurf, VSCodium and the other VS Code forks pull from Open VSX. The id is the same as on the Marketplace.
AI agents
npx -y decisionlint-mcpRuns DecisionLint's engine as an MCP server, so an agent can call lint_text with no editor involved. VS Code 1.101+ needs nothing — the extension registers it for you.
Channels
Where it ships
One engine, 4 places to get it. Copy the one for the editor you use.
MCP server
the same engine, callable by an agent with no editor in the loopnpx -y decisionlint-mcp
Links
Where DecisionLint lives
- VS Code Marketplacedecisionlint
- Open VSXfor Cursor, Windsurf and VSCodium
- npmdecisionlint
- GitHubsource, issues, changelog
- MCP registryio.github.nolindnaidoo/decisionlint
The family
The other 16
String-LEExtract every string in a codebase, so a person can read them
Numbers-LEExtract every hardcoded number in a codebase, so a person can check them
Paths-LEExtract every file path in a codebase, and say whether it still points at anything
Colors-LEExtract every color in a codebase, and say which ones are not in your palette
URLs-LEExtract every URL in a codebase, with its protocol and exact position
Dates-LEExtract every date and timestamp, and the exact instant each one resolves to
Units-LEExtract every quantity with its unit, normalized, and refuse the ambiguous ones by name
IDs-LEExtract every UUID, ULID, NanoID, ObjectId and Snowflake, and decode the time inside
IPs-LEExtract every IP address, CIDR block and MAC, normalized and classified by scope
Regex-LEFind every regex in a codebase, and report which can be driven into catastrophic backtracking
Scrape-LECheck whether a page is scrapeable before the scraper is written, and say when it cannot tell
Versions-LEFind where one dependency is constrained differently across a repository's manifests
i18n-LEIdentify the i18n library a project uses, then audit its catalogs by that library's rules
Secrets-LEFind hardcoded credentials in a codebase, and never print one into the report
EnvSync-LECompare the dotenv files in a tree, and say which keys are missing from which
Unicode-LEFind the Unicode that hides meaning — bidi controls, invisibles, homoglyphs, mixed scripts
