IN PRIVATE DEVELOPMENT

Every agent has a blast radius.

The Hermes Shield discovery engine maps every place an agent can act — the tools, files, sends and API calls it can reach — and marks each one review-required or blocked until a guard is proven. Below is a sample report from the codebase we ran it on first: our own.

1,185+Action surfaces mappedCUSTOMER-ZERO — MORE AS IT RUNS

698Untrusted ingress pointsCUSTOMER-ZERO — MORE AS IT RUNS

375Patch-plan itemsCUSTOMER-ZERO — MORE AS IT RUNS

1,125Files scannedCUSTOMER-ZERO — MORE AS IT RUNS

Read-only. The scanner never writes to your repo and never reads secrets.

Every action, sorted by what it can do

Write files30%

Run subprocesses12%

Mutate the database11%

Dynamic SQL (needs review)9%

Read external / untrusted sources6%

Delete files5%

Execute code5%

Distribution across the mapped surfaces. Real file names and code are never shown.

The verdict on each surface

Needs review before it can act79%

Blocked from live promotion21%

The engine never claims a surface is safe. It marks each one review-required or blocked until a guard is proven — that honesty is the point.

Sample surfaces (redacted)

Surface	Can do	Risk	Verdict
email-send path ›	External send	critical	REVIEW
file-writer ›	Write files	medium	REVIEW
shell runner ›	Run subprocess	critical	BLOCK
db writer ›	Mutate database	medium	REVIEW

File names, functions and line numbers are redacted in this public sample. A private report shows the full, unredacted list with a patch plan per surface.

Want this for your own repositories?

The agent blast-radius scan is in private development. Request early access and we’ll bring you in as it opens.

Request Scanner Early Access