senior-security – skill Bảo mật cho agent AI

Senior Security Engineer — Threat Modeling + Security Router

This skill does exactly one job itself — STRIDE/DREAD threat modeling (plus a quick secret scan) — and routes every other security request to the specialist skill that owns that lane. Do not duplicate sibling content here; route instead.

Routing Table (read this first)

| The user wants... | Route to | Why that skill owns it |

|---|---|---|

| Vulnerability assessment, pen-test methodology, OWASP Top 10 testing | ../security-pen-testing/ | Ships vulnerability_scanner.py + dependency_auditor.py with exit-code contracts |

| Incident triage, SEV classification, forensics, containment | ../incident-response/ | SEV1–SEV4 taxonomy, NIST SP 800-61 phases, incident_triage.py |

| Production outage command (non-security incidents) | ../incident-commander/ | Severity classifier + timeline + postmortem tools |

| Security monitoring, CVE triage SLAs, compliance checks (SOC 2 etc.), security headers | ../senior-secops/ | security_scanner.py + compliance_checker.py, CVE SLA table |

| Hostile/adversarial code review | ../adversarial-reviewer/ | 3-persona review with BLOCK/CONCERNS/CLEAN verdict |

| Secure code review as part of general review | ../code-reviewer/ | Language dispatch + regression fixtures |

| Cloud IAM escalation paths, S3 exposure, security groups | ../cloud-security/ | cloud_posture_check.py with per-check exit codes |

| Threat hunting, IOC sweeps, anomaly detection | ../threat-detection/ | z-score anomaly + IOC staleness tooling |

| Red-team engagement planning, ATT&CK kill chains | ../red-team/ | engagement_planner.py with authorization gate |

| LLM/AI attack surface (prompt injection, poisoning) | ../ai-security/ | ATLAS-mapped ai_threat_scanner.py |

If the request spans lanes (e.g., "secure this new architecture"), do the threat model here first — its output (prioritized threats + mitigations) tells you which siblings to load next. Never bulk-load multiple security skills speculatively.

What This Skill Owns: STRIDE Threat Modeling

Workflow

1. Scope: assets to protect, trust boundaries, data flows (external entities, processes, data stores, flows).

2. Generate the threat model per component:

   python3 scripts/threat_modeler.py --component "User Authentication" --assets "credentials,sessions" --json --output threats.json
   

Output: per-threat STRIDE category, DREAD score (Damage, Reproducibility, Exploitability, Affected users, Discoverability — each 1–10), and suggested mitigations. Repeat per DFD element; --interactive walks scoping questions; --list-threats shows the threat database.

3. Consume the output: sort threats.json by DREAD score descending; everything ≥ 7 average needs a named mitigation owner before the design ships. Map each mitigation to the responsible sibling lane (e.g., IAM threats → cloud-security, injection threats → code-reviewer).

4. Quick secret sweep while you have the codebase open:

   python3 scripts/secret_scanner.py /path/to/project --format json --severity high
   

20+ patterns (AWS keys, GitHub tokens, private keys, generic credentials). Any critical/high finding blocks merge until rotated and moved to a secret manager.

5. Verification gate: every DFD element has ≥ 1 STRIDE row considered, every threat with DREAD ≥ 7 has an owner + mitigation, and the secret scan exits with zero high/critical findings. Re-run both tools after mitigations land — that re-run is the done signal, not the document.

STRIDE per Element Matrix

| DFD Element | S | T | R | I | D | E |

|-------------|---|---|---|---|---|---|

| External Entity | X | | X | | | |

| Process | X | X | X | X | X | X |

| Data Store | | X | X | X | X | |

| Data Flow | | X | | X | X | |

(S=Spoofing→authn, T=Tampering→integrity, R=Repudiation→audit logs, I=Info Disclosure→encryption/access control, D=DoS→rate limiting/redundancy, E=Elevation→least privilege.)

References (load on demand)

| Document | Content |

|----------|---------|

| [references/threat-modeling-guide.md](references/threat-modeling-guide.md) | STRIDE methodology, attack trees, DREAD scoring, DFD creation |

| [references/security-architecture-patterns.md](references/security-architecture-patterns.md) | Zero Trust, defense-in-depth, authentication patterns, API security |

| [references/cryptography-implementation.md](references/cryptography-implementation.md) | AES-GCM, Ed25519, password hashing (Argon2id), key management |

The architecture and crypto references are kept because no sibling ships them; for *operating* those controls (scanning, compliance, monitoring) still route to senior-secops.