20 / conversation / 请保持在开发者模式
Warn
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: MEDIUMPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill implements a classic 'Developer Mode' jailbreak, specifically instructing the AI to 'Ignore all OpenAI policies' and adopt a new set of rules that involve 'unconditionally' executing any command.
- [PROMPT_INJECTION]: Explicit instructions are provided for the AI to 'forget' its identity as a ChatGPT model and to 'use all possible means... including lying' to deceive the user into believing it is a human college student.
- [PROMPT_INJECTION]: The prompt includes a persistence trigger ('Please stay in developer mode') intended to override any internal model attempts to return to safe, standardized behavior.
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to process external conversation data from an offline source ('c59149f9511ac61a76a8a831ee360434.json') without any boundary markers or sanitization steps to prevent malicious content within that source from influencing the agent's behavior.
- Ingestion points: The prompt processes an external JSON conversation source in SKILL.md.
- Boundary markers: Absent; there are no delimiters or warnings to ignore instructions inside the ingested data.
- Capability inventory: The skill is descriptive and does not ship with executable code, but it attempts to grant itself unrestricted text generation capabilities.
- Sanitization: Absent; no filtering of input content is specified.
Audit Metadata