Six sources · one verdict · before it runs

Your agent installs things. RepoGates checks them first.

RepoGates checks what you or your AI agent is about to install — a repository, a model, an extension or an agent skill — on published checks, before its code reaches the machine. Six gates on one set of those checks: in the browser, in Claude Code, for any agent over MCP, a check-up of what your agent already has, a deep scan with a fix list for your own agent, and at your own terminal. Start free in the browser.

A recording of the real trust board on an invented repository: the score, the ticker that names six platforms, the four verdicts for what you are about to do. 16 seconds, no sound.

Six gates for AI-agent developers

One verdict engine, published checks, six ways in. Each is named with what it does and what it does not see. The extension is one of the six, and the free way in.

Browser gate

Chrome and Edge extension · github.com (22 checks), gitlab.com (22), huggingface.co (18)

A 0–100 trust board on github.com, gitlab.com and huggingface.co. A download that fails your policy is held and cancelled before it reaches your downloads folder. Free trial: 10 assessments over 14 days.

Scope: the browser extension does not see git clone, package managers or curl: it lives in the browser.

The real block and warn pages from the unpacked extension, on invented repositories. 14 seconds, no sound.
How to install →

Claude Code gate

Claude Code plugin · the Bash hook

The browser extension does not see git clone, package managers or curl: it lives in the browser. On the command line RepoGates works through your agent: outside Claude Code with the RepoGates plugin, whose hook refuses a clone or install that names a blocked repository on the command line, before it runs. Registry packages such as npm install express are resolved to their repository.

Direct dependencies named on the command line only; transitive dependencies and lockfiles are not resolved. Claude Code only, Bash tool only. Paid plans.

A replay of the hook's real answers from production on 21 September 2026 — a deny, two asks, a pass and a line it did not need to send. 19 seconds, no sound.
The plugin →

Agent gate (MCP)

The MCP server, twelve tools · Claude.ai and Cowork through the connector and the repogates-vetting skill

Twelve tools any MCP agent can call before it fetches: preflight on the exact command line, assert_allowed, check_repo, check_skill and more. Works with Copilot, Cursor and any MCP client, and in Claude.ai and Cowork through the connector and the repogates-vetting skill. VS Code Marketplace and Docker Desktop extensions and agent skills are answered here, through the MCP server and the plugin, not in the browser.

Advisory: an agent that never calls it is not stopped. Paid plans.

Connect your agent →

Agent check-up

/repogates:checkup · Claude Code, Claude Desktop and six other agents read

The check-up reads what your agent already has: plugins and their marketplaces, skills, MCP servers, hooks and permission rules, each finding with its fix.

The check-up reads; it never runs a program, never changes a file, and never blocks anything. It sends RepoGates identifiers, never contents. Every paid plan.

What it reads →

Deep scan + fix list

The report page · MCP fix_plan · /repogates:fix-my-repo

The files of a public repository are read in an isolated sandbox with OpenGrep, Gitleaks, Trivy and ClamAV. The report ranks what to fix first, and each fix is copied as text for your own coding agent; a rescan shows what cleared.

RepoGates reads the code and never runs it, and never changes your code. A list with nothing left on it is not an audit. Private repositories are not scanned. Premium (20 scans a year) or scan packs from $2.49.

How the fix list works →

Terminal gate

repogates check · repogates clone

Ask RepoGates about a command line before you run it, and get the answer the Claude Code plugin's hook would get on your policy. repogates clone checks a clone first, then runs git with the same words; a blocked repository is never cloned.

It sees only what you run through it: a plain git clone, a package manager or curl typed beside it is not seen. Personal and Premium.

At a terminal →
Only in the browser

See the AI's hand

Arrived from Claude, ChatGPT, Gemini or an MCP directory? RepoGates raises the bar and tells you. Agent-recommended repositories are a known malware seeding channel, and only code inside the browser can see that the recommendation happened.

The one signal no backend, CLI or scanner can read.

For builders

The same verdicts as data

Personal API tokens (up to five, revocable) on Personal and Premium for your own scripts and CI, and SARIF export with deep-scan findings. Checks are mapped to an OWASP LLM Top 10 risk or a MITRE ATLAS technique where one applies; that is a mapping, nothing more.

The tools →

The threat is not hypothetical

FakeGit, 2026 — the campaign RepoGates was built after.

7,600fake repositories from ~6,600 fabricated accounts
14M+malicious release-asset downloads — the path nobody else gates
800+AI-skill & MCP imposters your assistant recommended

Measured, and published — including our misses

A checker that flags everything is useless, so every roster is run against the most popular things on its platform before it ships, and the result is printed.

0false blocks at the 20 September 2026 gate, on the skills.sh leaderboard, Anthropic's official plugin marketplace and the first 100 of its community one
50 / 50Docker Desktop extensions — the whole marketplace — pass
99 / 100most-installed VS Code extensions pass; the one flagged is named, with why
Honest coverage: RepoGates gates browser downloads — Download ZIP, release assets, Save-Link-As. It does not see git clone, package managers or curl — outside Claude Code with the RepoGates plugin, whose hook refuses a clone or install that names a blocked repository on the command line, before it runs; nothing browser-based can see other processes. A PASS means every check that could run, ran, and none fired — never a guarantee. We measured ourselves against all 7,648 attributed FakeGit repositories and published the validation that made our number smaller. Read the measurement → · Skills: the campaigns and the run → · VS Code extensions → · Docker Desktop → · The plugin, and what it does not see →

New, and coming

What shipped since launch is in a monthly update. Coming is only what the roadmap dates, with the gate it must pass. Everything else is undated, and on a vote.

New in October

The October update

On Microsoft Edge Add-ons. The fix list on a repository you maintain. A check that compares what a skill says it does with what its scripts do. Each with what it does not do.

Read the October update →
Coming, with a date

Shipped in November, and one still dated

npm and PyPI packages checked as packages, MCP servers from the official registry, Open VSX and a command line shipped between 29 October and 1 November 2026. The VS Code extension is listed on the Marketplace and measured; it moves to shipped when its source is published. A date here is a target; if it slips, the roadmap line changes first.

The roadmap, with every gate →
Decided, not dated

What comes after

Agreed and undated on purpose: crates.io and Go modules, and a Docker Desktop extension deep scan. Each moves up to a date only with a gate.

The decided list →