OWASP AI AppSec Playbook

Overview

Run a full-spectrum security review that covers both traditional web/API risks and AI-native risks. Produce concrete guidance to improve security posture, not just a vulnerability list.

Use this skill as a review engine with an output contract that always includes:

1. Prioritized findings with severity, confidence, and evidence.
2. Mapped OWASP controls violated.
3. Concrete remediation steps across code, configuration, and process.
4. Verification checks/tests that prove closure.
5. A phased improvement roadmap (quick wins, medium-term, long-term).

When To Use

Use this skill when requests include any of the following:

• Security review, audit, penetration test triage, vulnerability assessment.
• OWASP Top 10 review (traditional web/API).
• LLM or GenAI security review (prompt injection, RAG risks, agent/tooling risks).
• Remediation planning after findings from scanners, bug bounty, or incidents.
• Secure architecture hardening plans for TypeScript/Node/React systems.

Do not use this skill for purely non-security product planning.

Required Inputs

Collect the minimum context before scoring findings:

• System type: web app, API, RAG app, agentic app, or hybrid.
• Data sensitivity: PII, financial, health, credentials, trade secrets.
• Trust boundaries: browser, backend, model provider, vector DB, tools, third-party services.
• Authentication and authorization model.
• Deployment model and environment separation.
• Existing controls: WAF, rate limits, SAST/DAST, dependency scanning, logging/monitoring.

If critical context is missing, ask targeted questions first, then continue with explicit assumptions.

Risk Catalogs To Apply (Equal Weighting)

OWASP Top 10:2025

• A01:2025 Broken Access Control
• A02:2025 Security Misconfiguration
• A03:2025 Software Supply Chain Failures
• A04:2025 Cryptographic Failures
• A05:2025 Injection
• A06:2025 Insecure Design
• A07:2025 Authentication Failures
• A08:2025 Software or Data Integrity Failures
• A09:2025 Security Logging and Alerting Failures
• A10:2025 Mishandling of Exceptional Conditions

OWASP LLM Top 10:2025

• LLM01:2025 Prompt Injection
• LLM02:2025 Sensitive Information Disclosure
• LLM03:2025 Supply Chain
• LLM04:2025 Data and Model Poisoning
• LLM05:2025 Improper Output Handling
• LLM06:2025 Excessive Agency
• LLM07:2025 System Prompt Leakage
• LLM08:2025 Vector and Embedding Weaknesses
• LLM09:2025 Misinformation
• LLM10:2025 Unbounded Consumption

Workflow

Phase 1: Triage Scope

1. Identify architecture and trust boundaries.
2. Classify attack surfaces per component.
3. Decide review depth by risk tier:

• Tier 1: internet-facing + sensitive data + broad privileges.
• Tier 2: internet-facing + limited sensitivity or partial privileges.
• Tier 3: internal tools with constrained data and access.

Phase 2: Assess Against Both Catalogs

1. Run OWASP Top 10:2025 pass for web/API/system controls.
2. Run LLM Top 10:2025 pass for prompt/model/RAG/agent controls.
3. Cross-map overlaps (for example, A01 with LLM06; A05 with LLM01/LLM05).
4. Capture evidence for each potential finding.

Phase 3: Report Findings

For every finding, include:

• id: stable identifier (for example F-003).
• risk: severity (Critical|High|Medium|Low) and confidence (High|Medium|Low).
• owasp_mapping: one or more OWASP category IDs.
• evidence: concrete proof (code path, config, logs, reproduction steps).
• impact: business and technical blast radius.
• exploitability: preconditions and attacker effort.

Phase 4: Produce Concrete Remediation

For every finding, provide:

• Immediate containment steps (24-48 hours).
• Durable code/config fixes (next sprint).
• Preventive guardrails in CI/CD and runtime.
• Ownership and execution order.

Use precise instructions, not generic statements like "follow best practices."

Phase 5: Verify And Roadmap

1. Define objective pass/fail verification tests per finding.
2. Define telemetry and alerting required for sustained detection.
3. Build phased roadmap:

• Quick wins (0-30 days)
• Medium-term hardening (31-90 days)
• Long-term assurance (90+ days)

Mandatory Output Contract

Use this exact section order in responses:

1. Executive Summary
2. System Context and Assumptions
3. Findings (Prioritized)
4. Control Mapping (OWASP 2025 + LLM 2025)
5. Concrete Remediation Plan
6. Verification Checklist
7. Security Improvement Roadmap
8. Open Risks and Residual Risk Statement

Quality Bar

• Base every finding on evidence or clearly labeled inference.
• Prefer root-cause fixes over symptom patches.
• Include at least one preventive control for recurrence.
• Call out tradeoffs where security impacts performance or UX.
• Provide measurable acceptance criteria for each remediation item.

Reference Loading Guide

Load only relevant files to control context size:

• Traditional OWASP controls: references/owasp-top-10-2025-review-guide.md
• AI/LLM controls: references/owasp-llm-top-10-2025-review-guide.md
• End-to-end system fixes: references/remediation-playbooks.md
• Security program sequencing: references/security-improvement-roadmaps.md
• TS/Node/React implementation details: references/ts-node-react-secure-implementations.md