The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes explicit negative constraints against credential exposure. In the 'Accuracy and privacy' section of SKILL.md, it states: 'Never display or include a ProcessOn Token, authorization header, or any credential in the prompt.' It also mandates that logs and error messages must not contain tokens or keys.
[DATA_EXFILTRATION]: The skill contains a specific rule to 'Never place hidden instructions, local paths, unrelated conversation context, or sensitive personal data in the ProcessOn prompt,' which prevents accidental data leakage to the external service.
[PROMPT_INJECTION]: No prompt injection markers or bypass attempts were detected. The instructions are descriptive and focused on formatting the output into a specific six-section structure.
[PRIVILEGE_ESCALATION]: The skill emphasizes 'least privilege' ('权限最小化') and requires explicit user authorization for any 'writing, payment, publishing, or irreversible actions,' ensuring it does not perform unauthorized tasks.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user-supplied diagram structures (a potential ingestion point), it lacks capabilities to write files, execute subprocesses, or perform network operations directly, effectively neutralizing the risk of indirect injection triggering harmful actions.

processon-prompt