Imported from yknothing/prodcraft (
eval/02-architecture/pc-security-design/manual-run-2026-04-10-access-review/eval-1-access-review-modernization-security-design/skill-under-test/SKILL.md). Install upstream withnpx skills add yknothing/prodcraft --skill skill-under-test. Copyright stays with the author.
Security Design
Make the security model an architectural decision, not a bug-fixing exercise after implementation.
Context
Security design identifies what must be protected, who might abuse the system, and which controls must exist at each boundary. It sits upstream of security audit: the goal here is to design the defenses, not just inspect the code later.
In Prodcraft, security design is most valuable when the system adds new trust boundaries, handles sensitive data, or depends on brownfield coexistence where old and new controls may differ.
Inputs
- architecture-doc -- Defines the system boundaries, deployment topology, and interaction patterns.
- api-contract -- Identifies externally visible actions, data entry points, and policy-sensitive operations.
Process
Step 1: Mark Assets and Trust Boundaries
Identify the data, operations, and integrations that would matter to an attacker or regulator. Draw the trust boundaries explicitly: browser to backend, service to service, job worker to datastore, admin surface to normal user surface.
Step 2: Enumerate Abuse Paths
For each boundary, ask:
- how could an unauthorized actor reach this path
- what happens if input is malicious, replayed, or oversized
- what data could leak through logs, errors, or side channels
- how could rollout or coexistence create a weaker path than the intended one
Step 3: Assign Concrete Controls
Define the required controls at each boundary:
- authentication and session controls
- authorization and tenant isolation
- input validation and output encoding
- secret handling and key rotation expectations
- audit logging, alerting, and fail-closed behavior
Step 4: Record the Threat Model
Capture attacker assumptions, control decisions, unresolved risks, and the checks downstream phases must enforce. The output should be actionable enough that implementation and security audit know what to build and what to verify.
Outputs
- threat-model -- Assets, trust boundaries, abuse paths, required controls, and explicit residual risks.
Quality Gate
- Trust boundaries are explicit
- High-value assets and abuse paths are documented
- Required controls are assigned per boundary
- Logging and fail-closed expectations are clear
- Residual risks and unresolved assumptions are recorded for audit or follow-up
Anti-Patterns
- Security by generic checklist -- controls that ignore the real boundaries of the system.
- Auth-only thinking -- focusing on login while ignoring authorization, secret handling, or data leakage.
- Invisible residual risk -- acting as if security is complete when key assumptions remain open.
- Brownfield blind spot -- designing new controls without modeling weaker legacy paths still in service.
Related Skills
- system-design -- supplies the structural boundaries to secure
- security-audit -- verifies the implementation against this design
- deployment-strategy -- uses the threat model when rollout posture affects exposure
- monitoring-observability -- exposes the signals needed to detect security-relevant failure modes
