Commands and launchers
Hooks, shell grants, uvx and npx paths.
AI is more dangerous than you think
26.1% of 31,132 agent skills analyzed in the wild contained at least one vulnerability. Source: Agent Skills in the Wild
Offline-first, precision security linter for SKILLS, MCP, plugins, configs and other AI infrastructure.
Quick run
npx lintai-cli scan .Runs locally
Agent config
JSON + SARIF
Local scanRuns locally
What lintai checks locally
It scans the agent files you already use.
$ lintai scan .Tool and MCP access
Permissions, MCP servers, connected tools.
Shared defaults
Team settings, rules, and instruction files.
Example finding
.claude/settings.jsonBroad shell access in shared settings
This shared config grants more shell power than most local setups need.
What local agent config hides
These files look routine, but they quietly decide what a local agent can run, access, and inherit.
Broad shell access in shared settings
A small permissions shortcut can quietly widen shell authority.
Looks harmless
A harmless shared settings block.
Actually changes
Turns team defaults into broader shell access than the repo needs.
Dynamic package launch in mcp.json
A routine command path can quietly change what code the agent runs.
Looks harmless
Normal mcp.json wiring or setup guidance.
Actually changes
Downloads or resolves agent tooling dynamically at runtime.
Hook command with no clear boundary
A shared hook can run with no clear execution guardrails.
Looks harmless
Simple setup glue in a repo hook.
Actually changes
Executes commands with unclear time, tool, or boundary limits.
What lintai catches in practice
These are typical things lintai highlights in committed config, shared settings, and editor instructions.
Why run lintai locally first
So risky defaults are visible on your machine before they quietly become normal team config.
Runs on your machine
Built for local runs and CI
Without a dedicated check
Depends on what a person notices
Shows the exact reason for each finding
Stable ids with structured evidence
Without a dedicated check
Explanations vary by person
Separates release-grade checks from preview signals
Documented in the shipped product posture
Without a dedicated check
Usually lives in tribal knowledge
Checks agent config people usually trust by default
Skills, MCP, hooks, settings, instruction files
Without a dedicated check
Easy to miss harmless-looking files
Also useful later
Fits into CI and code scanning
Text, JSON, and SARIF are built in
Can inspect what is already installed locally
scan-known and inventory-os are built in
Run your first scan
Pick the fastest supported path and run lintai locally before you wire in anything else.
Latest release · v0.1.1 · May 28, 2026
Fast path to the first scan
The fastest supported path to a real local lintai scan.
Choose a run path
Fastest supported path on macOS and Linux. Downloads the verified release installer, installs lintai, and immediately scans the current repository.
01
Run the first repo scan
Command
npx lintai-cli scan .Best default for a first local check when you want the shortest real path to findings.
02
Export SARIF for CI or code scanning
Command
lintai scan . --format sarifUse SARIF when you want the same scan integrated into CI and downstream tooling.
03
Inspect resolved policy when you add config
Command
lintai explain-config lintai.tomlUseful once the target repo has a local lintai policy and you want to confirm the active preset and rule posture.
What to know before CI
The short honest version: where the current release is strong and where its trust boundary still sits.
Repository-local scan surface
Initial public releaseThe current product story is the repo-local scan path for AI-native files such as skills, MCP configs, hook settings, and local client policy surfaces.
Stable vs Preview policy
Documented nowStable findings are the release-quality baseline. Preview remains useful but explicitly non-baseline and more context-sensitive.
Offline advisory lane
Opt-inDependency advisory matching is intentionally opt-in and limited to committed npm lockfiles against the active offline snapshot.
Installed artifact audit
Available nowscan-known, inventory-os, and policy-os extend lintai beyond repo scans when you need to inspect what local AI clients already have configured.
Got questions? We've got answers
The practical version: what lintai scans, where the current v0.1 boundary is, and how teams usually start using it.
Start here
Need a clearer next step?
Most teams only need three pages first: the docs index, the positioning note, and the release contract.
Open docs indexBrowse the rules and preset reference without leaving the repo context.
Open positioning noteSee who lintai is for, what it catches, and what it does not promise yet.
Check release contractRead the current public release posture before promising more than the product ships.