Attackers will always find a way in. Get to know your blast radius, so you can ensure there's nothing of consequence for them to find.
Every record sits in the clear behind one credential. One read takes the lot.
Works where you code
No pipeline, no repo permissions, no new dashboard. Connect Raziel, Tide's MCP, to your AI coding agent and ask Map my blast radius. Your agent reads your actual source and writes the assessment in the same chat.
One click or one command in the coding agent you already use.
Prompt Map my blast radius
A Blast Radius Assessment of your project, every finding scored by what it yields, with the evidence cited.
A threat model, not a linter
Nobody ships perfect code, stays fully patched or controls every dependency. Conventional scanners look for known bad patterns, CVEs and misconfigurations. All worth fixing, but never enough, because your attacker will get in through a path you're not aware of or one that's out of your control.
The Blast Radius scan assumes they're already inside, or an insider, and maps what they'd find. The places where they can commandeer the authority to cause catastrophic damage. Remove those, and a break-in becomes inconsequential.
Every finding is rated on the four-tier blast radius scale, and any relevant OWASP, CWE and severity references so it maps to the reviews you already run.
One compromise yields the whole platform.
A trust root or a whole class, one step removed.
A bounded set. One record, one session, one user.
A request-edge or config weakness that is not itself a central point.
Sample findings
This is the scan's actual output for MedTrack Pro, a vibe-coded health tracker built on Next.js, Clerk, Prisma and Gemini. Exactly the kind of app AI helps you ship in a weekend, and exactly where the scan earns its keep. The heart of the report is the single-point map. Reach a point on the left and you hold everything on the right.
| Reach this one point | Blast radius | And you hold |
|---|---|---|
| One read of the PostgreSQL database | TOTAL | every user's full symptom history, in clear, at once. Symptom types, severities, and free-text notes (F-02) |
| The external identity provider (or a false verdict from it) | SYSTEMIC | the ability to be any user, or to have a false authentication accepted with no way for the app to detect it (F-01) |
| A live session token | CONTAINED | one user's session, replayable from anywhere until it expires (F-04) |
| The AI-summary path | CONTAINED | one user's clinical notes, disclosed to a third-party AI provider (F-03) |
| The middleware error path | LIMITED | a bypass of the edge auth gate, bounded today by the per-route checks behind it (F-05) |
“The findings below are not a list of broken checks. They are the points where authority is concentrated by design, and they would remain after every checkable bug was fixed.”
The full report records a verdict for every surface it examined, including the ones that held. MedTrack's per-user authorization came back SOUND, with the file-and-line evidence cited. The scan names what your app gets right as plainly as what it doesn't.
Then remediate
When the scan completes, ask your agent for the remediation report. For each finding it shows what changes when you build on TideCloak, the mechanism that does it, and what stays in your hands. Findings marked NOT CHANGED are ordinary app fixes, handed back to you with the fix spelled out. From there, the same agent can implement the changes, guided step by step by Raziel's playbooks.
F-01Identity in one external provider, trusted blindly
PRISM + threshold-signed identity
F-02All symptom records server-readable at rest
Hermetic E2EE
F-04Unbound bearer session
DPoP binding
F-03Symptom notes sent to a third-party AI
App data-governance decision
F-05Auth middleware fails open
App error handling (fail closed)
From the remediation report for MedTrack Pro. Three findings removed outright, two returned to the app with concrete fixes.
The first question any security-minded dev asks of a security scanner is who else gets to see the code. Here, only your agent does.
The scan is performed by the coding agent you already trust with your source, locally. Raziel supplies the methodology; it never sees your code.
Every tool Raziel exposes serves guidance. It cannot touch your machine, your repository or your data.
Nothing uploads and no SaaS scanner asks for access to your source. The report exists only where your agent wrote it.
Add Raziel, open your project, and ask.