← ClaudeAtlas

arch-security-reviewlisted

Analyzes an uploaded architecture diagram (image/screenshot/exported drawing) and produces a security architecture review mapped against PCI DSS, PCI SSF, ISO 27001, NIST CSF, NIST 800-207 Zero Trust, CIS Controls v8, CSA CCM, OWASP ASVS/API Security Top 10, STRIDE, RBI cybersecurity/digital-payment regulations, and India's DPDP Act/Rules (privacy). Trigger whenever the user attaches/references an architecture, data-flow, or network diagram and asks for a security review, threat model, compliance gap analysis, or standards assessment (PCI/ISO/NIST/CIS/RBI/DPDP/privacy/cloud/zero-trust/API security) — even without naming a specific standard. Also trigger for reviewing a design before it's built, cardholder-data/payment/personal-data flows, segmentation/trust boundaries, or a bank/NBFC/payment-system/consumer-app design for Indian regulatory alignment.
sharmajii10/arch-security-review · ★ 0 · Code & Development · score 62
Install: claude install-skill sharmajii10/arch-security-review
# Architecture Diagram Security Review ## What this does Takes one architecture diagram (a picture — PNG/JPG screenshot, exported draw.io/Lucidchart/Visio image, whiteboard photo, etc.) and turns it into a structured security review: an inventory of what's in the diagram, the trust boundaries and data flows, a STRIDE threat pass, and findings mapped against whichever of ten standards actually apply — PCI DSS, PCI SSF, ISO 27001, NIST CSF, NIST 800-207 (Zero Trust), CIS Controls v8, CSA CCM, OWASP ASVS/API Security Top 10, and RBI's cybersecurity/digital-payment regulations — each finding with a concrete recommendation. ## Why the process matters A security architecture review is only useful if the reader can trust it. Two failure modes are equally bad: - **Hallucinated confidence** — inventing a control that "must" be there because a well-architected system would have it, or inferring encryption/auth that isn't actually shown. This creates false assurance, which is worse than no review at all. - **Noise** — flagging every theoretically possible issue regardless of what's actually visible, burying the findings that matter under boilerplate. The fix for both is the same discipline: only claim what the diagram shows or explicitly labels, mark everything else as an open question, and prioritize findings by actual exposure (is this component internet-facing, does it touch cardholder/PII data, does it cross a trust boundary) rather than by how many standards a generic issue ha