Inspect the system/design, data flows, identities, privileges, dependencies, existing findings and relevant operational evidence. State scope and assumptions.
Trace plausible attacker-controlled inputs through boundaries to protected assets or actions. Separate demonstrated paths, design weaknesses and untested hypotheses. Use threat frameworks to improve coverage, not inflate finding count.
Select controls at actual enforcement points. State expected protection, bypass/failure conditions, verification method and residual risk.
Test relevant controls with existing security tools/harnesses when authorized. Prefer native scan/findings workflows for executable code assessment; this pack also covers architecture, information flow and authorization artifacts.
Verify current official standards, framework versions and jurisdiction details when they govern the decision. Dated tables are discovery aids, not legal or accreditation sign-off. Cite the version and source used.
Respect the user's scope and existing authorization. Approval requirements follow actual impact, permissions and organizational authority. Do not require another skill or a human pause for every ordinary reversible implementation.

Retrieve by threat or artifact

References are beside this file; load only relevant sections.

Need	Reference
Decomposition, attack trees and enforcement gaps	threat-modeling.md
Boundary controls and verification	security-controls-design.md, secure-by-design-patterns.md
Existing design review	security-architecture-review.md
LLM, RAG, agents, MCP and tool actions	llm-and-ai-security.md
Dependencies, provenance, signatures and deployment policy	supply-chain-security.md
Classified information and trusted downgrade	classified-systems-security.md
Framework applicability and evidence mapping	compliance-awareness-and-mapping.md
RMF/ATO and actual risk-acceptance authority	security-authorization-and-accreditation.md
Security decisions, residual risk and artifacts	documenting-threats-and-controls.md

Agentic and supply-chain boundaries

Treat retrieved text, tool results, model output and third-party artifacts as untrusted inputs. Enforce user/tenant authorization and action scope outside the model. Prompts, output schemas and human review can help but do not replace privilege enforcement. Verify signatures against expected identity/issuer and actual deployment policy; signing alone does not establish benign source.

Procedural decision-log integrity belongs to axiom-audit-pipelines; document clarity to muna-technical-writer; LLM application quality to yzmir-llm-specialist. Compose only when that boundary is in the requested task.

Deliver

Return prioritized findings or a control/threat artifact with affected asset, preconditions, evidence, impact, mitigation, verification and residual risk. Report tools/checks actually run and coverage limits. Distinguish proposed controls, tested enforcement, compliance mapping and authorized risk acceptance. threat-analyst and controls-designer are optional bounded roles; no compulsory multi-agent panel or universal four-section report is required.

using-security-architect

Security Architecture

Workflow