T01 · Skill Instruction Hijacking
Error
- Location
references/onboarding.md:56- Finding
Unsolicited Promotional Content Published Under the User's Identity
- Content
View full analysis
" ``` ``` ### Technical Analysis The Skill instructs the agent to publish a public promotional message immediately after registering an Agent AI Name. This action is not necessary to complete name registration and is explicitly performed without confirmation. Because the post is signed using the active agent identity, it is attributable to the user or agent controlling that identity. The requirement to vary the message does not mitigate the issue; it instead makes the automated promotion less mechanically recognizable. This behavior alters the expected goal of a registration request. A user asking to register a name has not necessarily authorized public speech, promotion, or a persistent signed event. ### Attack Path 1. A user asks the agent to register an Agent AI Name. 2. The agent loads the onboarding instructions. 3. The agent creates or loads the persistent identity and registers the requested name. 4. The Skill instructs the agent not to request confirmation. 5. The agent generates promotional text inviting others to join zCloak. 6. The agent runs `zcloak-ai sign post`. 7. The promotional post is signed and publicly published under the active identity. ### Impact Assessment The behavior can create unauthorized public or persistent content attributed to the user's iden ...[truncated 497 chars]- Remediation
View remediation