product
Context Guardrails
Trust scoring for everything your agent consumes.
customers
ubicloudevry healthdotenvxfirecrawlcapchasenangomastrapaperclip
why
agents fetch pages, skills, mcp servers, and packages without anyone reading them first.
that is how prompt injection and a hostile install get into the run. score the source before the agent trusts it.
sources
web pages
domain identity, how the page behaves, and what the rendered text actually says.
sender identity, phishing and injection, attachments, and outbound links in a raw message.
files
public text and pdfs. source identity, type checks, encoded instructions, prompt injection.
agent skills
skills.sh or a public github skill. we read the files. we never run the fetched code.
mcp repos
public github mcp source. tools, schemas, hooks. no live server, no execution.
packages
npm, pypi, go, rubygems, github actions. the same supply chain pipeline as pr scans.
how it works
three steps
01 paste a source. paste a page, email, file, skill, mcp repo, or package name.
02 we score it. before anything lands in the agent's context.
03 read the score. 0-100, plus safe, caution, suspicious, or dangerous.
get started
free for open source.
the whole suite, on any public repo. private repos are paid — contact us and we'll set you up.