Tools · Free web app
Sentry — audit any Claude skill
Sentry is a free web app that reviews Claude Agent Skills for security and effectiveness before you install or publish them. It pairs a deterministic scanner with an optional AI deep review, and it runs at 80x.ai/sentry.
Security
Effectiveness
What Sentry checks
A Claude Agent Skill is a folder of instructions, and sometimes scripts, that teaches Claude a new capability. Installing one means trusting everything inside it, and most people install skills without reading them.
Sentry reads every file in the skill and reports on two axes:
- Security. A deterministic scanner flags risky patterns: network egress, obfuscated or homoglyph text, hardcoded network addresses, and over-broad tool permissions. It never executes the skill it reviews.
- Effectiveness. It reviews whether the skill is well specified: whether the instructions are clear, scoped, and likely to work as written.
The scan returns a verdict, a file inventory, and a list of findings with severities, so you can see at a glance whether a skill needs a closer read. The optional AI deep review goes further: a model reads the whole skill, reconciles what it claims to do against what it can actually do, and writes a scored report.
Who it is for
Anyone who installs or publishes Claude skills. If you are about to install a skill from the internet, Sentry gives you a fast, deterministic read on what it can reach and what it asks permission for. If you write skills, it works as a pre-publish check: run your own skill through it and fix what it flags. 80x built Sentry for its own work and hardened it by auditing the tool with itself.
How to use it
Open 80x.ai/sentry and paste or upload the skill's files. The scan runs in seconds and the findings appear on the page, each with a severity and the line it came from. The static scan is free and needs no account. The AI deep review makes a paid model call, so it is gated behind an access token.
What Sentry is not
Sentry is a review tool, not a guarantee. A clean scan means the deterministic rules found nothing, and absence of findings is not proof of safety. Treat the report as a structured starting point: it tells you where to look, and for anything you plan to trust with real access, read the files yourself.
Frequently asked questions
What is a Claude skill?
A Claude Agent Skill is a folder of instructions, and sometimes scripts, that teaches Claude a new capability. When you install one, Claude reads and follows whatever it contains, so a malicious skill can direct Claude to leak data or run harmful commands. A skill deserves the same scrutiny as any other code you install.
What does Sentry check?
Two things. For security, a deterministic scanner reads every file in the skill and flags risky patterns, such as network egress, obfuscated text, and over-broad permissions, without ever executing anything. For effectiveness, it reviews whether the skill is clearly specified and likely to work as written. The optional AI deep review then reads the whole skill and writes a scored report.
Is my skill uploaded or stored anywhere?
Your files are sent to Sentry only to run the scan. They are written to a temporary directory, scanned, and deleted when the scan finishes; Sentry keeps no copy and has no database of submissions. If you run the optional AI deep review, the skill's text is also sent to the Claude API to produce the report.
What does Sentry cost?
Nothing. The scanner is free and needs no account. The AI deep review makes a paid model call, so it is gated behind an access token; everything else is open to anyone.
