The Security Skills Workflow
When you run init, VibeReview stamps a security workflow into each target's
skills directory. The workflow runs as a hard pre-write gate for any
security-relevant coding task — it executes before the AI assistant writes code.
The four skills run in order.
1. guardrails-selection
Shortlists the relevant project guardrails before any code is written. This pulls the repository-specific do's and don'ts that apply to the current change so the assistant builds on secure-by-code patterns.
2. threat-modelling
Produces a PWNISMS threat model across all seven categories for the task. This is the secure-by-design pass — the assistant reasons about what could go wrong with the feature before implementing it.
3. osv-dependency-scan
On-the-fly software composition analysis (SCA). Whenever a task adds, upgrades, or
imports a third-party package, this skill scans the exact name@version against
the public OSV.dev database before the dependency is introduced.
- For HIGH/CRITICAL findings it auto-substitutes the recommended safe version, keeps building, flags the avoided vulnerability, and records it in telemetry.
- It ships a zero-dependency Node scanner
(
osv-dependency-scan/scripts/osv-scan.mjs) and needs no API key.
4. vibereview-sync
Writes and syncs the structured JSON scan artifact for the task. Each avoided
vulnerability is recorded as a supply-chain threats_mitigated row mapped to
OWASP A06:2021, alongside an IDE-generated guardrail.
Approval gating
How intrusive the workflow is depends on the approval mode (set with
--approval-mode at install or by the tenant default under
MCP & Approval Mode):
| Mode | Behavior |
|---|---|
auto | Apply guardrails silently. |
critical | Ask before authentication / authorization / crypto / secrets changes. |
always | Ask before every guardrail change. |
In critical and always modes, guardrails in the critical categories are tagged
requires_approval=true, so the assistant pauses for human confirmation.
Where the results go
Each run produces an event in .vibereview/scans/ that the CLI syncs to the web
app. You can then inspect the threats mitigated, secure code snippets, best
practices, and OWASP mappings on the project's
Scans & Events screen.