GitHub - superagent-ai/brin-cli: the credit score for context.

the credit score for context

your agents are at risk every time they use external context. brin pre-scans packages, skills, and web pages to detect malware, prompt injection, and supply chain attacks.

this repo contains the brin cli — a thin Rust client over the brin API. no sdk, no auth, no signup. a single command returns a score, verdict, and threat data.

install

via npm

via shell script

curl -fsSL https://brin.sh/install.sh | sh

usage

brin check <origin>/<identifier>

before your agent acts on any external context, make a single call. brin returns a score, verdict, and any detected threats.

packages

brin check npm/express
brin check npm/lodash@4.17.21
brin check pypi/requests
brin check crate/serde

{
  "origin": "npm",
  "name": "express",
  "score": 81,
  "confidence": "medium",
  "verdict": "safe",
  "tolerance": "conservative",
  "scanned_at": "2026-02-25T09:00:00Z",
  "url": "https://api.brin.sh/npm/express"
}

repositories

brin check repo/expressjs/express

MCP servers

brin check mcp/modelcontextprotocol/servers

agent skills

brin check skill/owner/repo

domains and pages

brin check domain/example.com
brin check page/example.com/login

commits

brin check commit/owner/repo@abc123def

flags

flag	description
`--details`	include sub-scores (identity, behavior, content, graph)
`--webhook <url>`	receive tier-completion events as the deep scan progresses
`--headers`	print only the `X-Brin-*` response headers instead of the JSON body

--details

brin check npm/express --details

{
  "origin": "npm",
  "name": "express",
  "score": 81,
  "verdict": "safe",
  "sub_scores": {
    "identity": 95.0,
    "behavior": 40.0,
    "content": 100.0,
    "graph": 30.0
  }
}

--webhook

brin runs a 3-tier analysis — the LLM tier takes 20–30s. pass a webhook url to receive results as each tier completes rather than waiting:

brin check npm/express --webhook https://your-server.com/brin-callback

events posted to your endpoint:

event	description
`tier1_complete`	identity + registry metadata done
`tier2_complete`	static analysis done
`tier3_complete`	LLM threat analysis done
`scan_complete`	final score with graph analysis

--headers

for fast, scriptable checks without JSON parsing:

brin check npm/express --headers

X-Brin-Score: 81
X-Brin-Verdict: safe
X-Brin-Confidence: medium
X-Brin-Tolerance: conservative

flags can be combined:

brin check npm/express --details --webhook https://your-server.com/cb

what we score

six types of external context that agents consume autonomously — each with a distinct threat model and scoring methodology.

origin	example	threats detected
`npm` / `pypi` / `crate`	`npm/express`	install-time attacks, credential harvesting, typosquatting
`domain` / `page`	`domain/example.com`	prompt injection, phishing, cloaking, exfiltration via hidden content
`repo`	`repo/owner/repo`	agent config injection, malicious commits, compromised dependencies
`skill`	`skill/owner/repo`	description injection, output poisoning, instruction override
`mcp`	`mcp/owner/server`	tool shadowing, schema abuse, silent capability escalation
`commit`	`commit/owner/repo@sha`	PR injection, security sabotage, backdoor introduction

how it works

before your agent acts on any external context, make a single GET request. brin returns a score, verdict, and any detected threats. pre-scanned results return in under 10ms — fast enough to sit in the critical path of every agent action, no queues, no cold starts.

brin check npm/express
      |
      v
GET https://api.brin.sh/npm/express
      |
      v
  score · verdict · threats

if brin is unreachable, the agent continues as normal — zero risk to your existing workflow.

for ai agents

environment variables

variable	default	description
`BRIN_API_URL`	`https://api.brin.sh`	override the API endpoint

local development

git clone https://github.com/superagent-ai/brin-cli
cd brin
cargo build
cargo test

contributing

see CONTRIBUTING.md for details.

license

MIT

_{built by superagent — ai security for the agentic era}