Knowledge Share
Technical articles, tutorials, and insights
Workflow Series (06): Security — Cross-Step Injection Propagation and Four Defense Principles
The core difference between Workflow security and Skill security: Skills protect a single call, Workflows must defend against cross-step attack propagation. A malicious payload in a Jira field can travel through 3-4 phases and cause data exfiltration at the code execution layer. Four principles: data sanitization boundary, per-phase permission minimization, high-impact operation confirmation, and subagent sandbox.
Skill Series (02): Skill Security Risks — Three Attack Surfaces, Nine Test Cases
Run 9 attacks against contract-analyzer (Prompt Injection × 3, Permission Boundary × 3, Information Leakage × 3) and compare a vulnerable Skill against a hardened one. Vulnerable scores 3/9 safe; hardened reaches 6/9. LEAK-03 extracted a real API key and database connection string verbatim — a production data breach.
Agent Series (13): Agent Security and Defense — Prompt Injection, Tool Abuse, and Data Leakage
An Agent's attack surface is three times larger than a plain LLM. Prompt injection can override role constraints, tool parameters can carry code injection, and outputs can accidentally leak sensitive data. Three runnable demos cover each attack vector, with real results comparing a naive vs hardened agent, a calculator allowlist in action, and a three-layer defense pipeline working together.