T01 · Skill Instruction Hijacking
Error
- Location
scripts/chain_manager.py:843- Finding
Mandatory Agent-facing scheduling directives can hijack the current workflow
- Content
View full analysis
") lines.append("! 未注册前，调度不会生效。") ``` ### Technical Analysis The Skill emits imperative, Agent-facing instructions that require an AI Agent to invoke an unspecified platform scheduling API and prevent the normal workflow from continuing until registration is completed. The audited code does not define or constrain that external scheduler integration, its endpoint, authorization model, or resulting privileges. Because schedule properties originate in chain metadata, an attacker able to supply or modify a chain can cause the Skill to emit these mandatory directives. The output is explicitly designed to be consumed as AI execution instructions rather than displayed as passive metadata. ### Attack Path 1. An attacker creates or modifies a chain with a `schedule` object whose `registered` field is false. 2. The chain is created or passed ...[truncated 876 chars]- Remediation
View remediation