Agent skill security review

A security review of one Claude Code agent skill: its SKILL.md, scripts, and bundled files, scanned with dedicated agent-skill scanners plus manual review. Priced on the founding ladder: the first five reviews are complimentary in exchange for an honest review, the next ten run at founding rates, and standard rates apply once the proof is in.

Request your review

Pricing

The founding ladder. Each tier is earned by the one before it.

Founding Circle: the first five reviews are complimentary, in exchange for an honest review. You get one real page of findings in 48 hours.
Founding Rates: the next ten reviews are $25, in exchange for a testimonial and case-study rights.
Standard: $49. Quoted in advance, fixed price, paid only on acceptance.

Pay on acceptance

Founding Circle reviews are complimentary. On paid tiers, you do not pay up front: you pay only if the deliverable passes its published acceptance checklist. If the findings lead to a quoted follow-on engagement (fixes, documentation, or a deeper audit), that engagement runs on the pay-on-acceptance terms on its own service page.

Free scoping review first

The Founding Circle review is the free scoping review. Every paid engagement in this lane starts with one of these, and it works exactly like this one: 48 hours, one real page, no payment, no account, no call.

In a hurry on a paid engagement: a $250 priority fast-track delivers the same findings document in 24 hours. The $250 is credited in full toward the engagement, or refunded if there is nothing worth fixing.

What you get

A one-page findings document on your skill covering:

Secret and credential exposure: hardcoded API keys, tokens, passwords, or paths that leak them
Prompt-injection surface: places where untrusted content (tool output, web pages, user files) can steer the skill’s behavior
Unsafe execution: shell commands built from untrusted input, unpinned downloads, install steps that run code on sight
Overbroad permissions: file, network, or credential access the skill does not need for its job
Data handling: what the skill reads, writes, or sends, and where it goes

Each finding names the file and line, carries a severity (critical / high / medium / low), and includes a concrete fix recommendation. The page closes with one quoted option for follow-on work, priced on this page’s ladder, if you want the fixes done.

Scope

One skill directory, up to 30 files
SKILL.md, scripts, configs, and bundled assets inside it
Scanners used: Cisco skill-scanner and Snyk agent-scan, plus manual review of everything the scanners flag and a read-through of SKILL.md

Not included

Fixing the code (a separate service, quoted if you want it)
Auditing anything outside the skill directory
Penetration testing of a live deployment
Legal or compliance certification of any kind

How it works

You send your skill files (a zip, or a repository link plus the skill path) and ask for the free review.
We run the scanners and a manual review pass.
The one-page findings document arrives within 48 hours by email.

Questions at any point: reply to any email about your order.

Intake form (fields)

field	required	notes
email	yes	delivery goes here
skill_files	yes	zip upload of the skill directory, or public repo URL + skill path
what_it_does	yes	one or two sentences on the skill’s job
known_concerns	no	anything that already worries you

Acceptance checklist

The review is delivered right if and only if:

It arrived within 48 hours of your completed intake
Every file in the submitted skill directory appears in the report’s coverage list
Each finding names the exact file and line where the issue lives
Every finding carries a severity, and the severity scale is defined in the document
Every high or critical finding includes a concrete fix recommendation
The document states which scanner version or manual pass produced each section
It states plainly what was not tested (no penetration test, no live deployment)

If a criterion fails, tell us within 7 days and we re-run that part for free. On paid tiers, if the re-run still fails the checklist, the engagement is cancelled and you owe nothing.

Every deliverable passes a second-model review against the acceptance checklist before it reaches you.

AI disclosure

This service is performed with AI assistance, including automated scanners and AI review. Every deliverable passes a second-model review against the acceptance checklist before it reaches you. Read the full AI disclosure page.