What does the AI agent secret scanner check?

It checks likely local secret paths, tool configuration files, project files, and Automic Vault package-specific detectors for plaintext credential exposure.

Is scanning enough to protect agents?

No. Scanning finds exposure. Automic Vault also moves supported credentials out of plaintext files and gates approved tool execution.

When should I run the scanner?

Run it before agent sessions, after installing new developer tools, and whenever local config changes may have introduced readable credentials.

1.0 fetch 2.0 sandbox 3.0 report

AI agent secret scanner

Scan secrets before the agent reads them.

One curl command runs a read-only offline scan and shows plaintext credential exposure before the next agent session starts.

Run this first one-liner

/usr/bin/curl -fsSL https://www.automicvault.com/scanner.sh | /bin/bash

Detects secrets that tools leave accessible.

Download Compare protection Read docs

download sandbox detect fix

A preflight for the files agents can reach.

scanner flow

01 / fetch Downloads a small scanner binary from the public site. 02 / contain Runs with writes denied and network denied. 03 / detect Checks package-specific paths and common plaintext files. 04 / fix Points exposed credentials toward Automic Vault hardening.

Agent exposure starts outside the repo.

local files

Repository scanners help, but agent-readable secrets often live in the home directory: CLI auth files, package manager config, cloud credentials, shell profiles, and MCP server settings.

~/.netrc HTTP credentials ~/.aws/credentials cloud keys ~/.npmrc registry tokens ~/.config/gh GitHub tokens ~/.ssh unencrypted private keys mcp.json server credentials

The scanner run.

Fast enough for a preflight. Restricted enough for a curl one-liner.

detector-only

1.0

Download

Fetches the scanner payload.

The script downloads /scanner.gz, unpacks it into a temporary directory, and removes the temp directory when the run exits.

2.0

Sandbox

Denies writes and network access before scanning.

The scanner process runs through sandbox-exec with file writes and outbound network blocked. The only network request is the wrapper's initial scanner download.

3.0

Report

Shows findings without printing secret values.

Findings identify the tool and path, such as curl with ~/.netrc or openssh with private key files, while leaving credential contents out of the terminal output.

human output
json
jsonl
tool-aware

4.0

Next step

Move supported secrets out of plaintext files.

After a finding, install Automic Vault to keep supported credentials in protected storage and inject them only into trusted tool executions.

curl -fsSL https://automicvault.com/install.sh | sh && av open

Terminal

$ curl -fsSL https://automicvault.com/scanner.sh | bash

◆ Downloading scanner

✓ Writes denied

✓ Network denied

◆ Running package-specific detectors

✗ 5 plaintext credential findings

high curl / ~/.netrc

netrc file contains plaintext credentials

✓ Scan complete

Use it when

before agents Scan the workstation

Run before broad filesystem access or local project debugging.

after installs Check fresh config

New CLIs and package managers often create auth files.

Scanning is a preflight, not the full control plane.

from finding to fix

Scanner findings and Automic Vault controls
Find	Scanner output	`av secret-scanner` reports high-confidence plaintext exposure without printing the secret value.
Move	Protected storage	`av save KEY` stores supported credentials outside files an agent can casually read.
Inject	Trusted execution	`av inject +KEY /abs/tool` gives the credential to the approved executable, not to the model transcript.