The Agent Skills Directory

[SAFE]: The skill defines a purely analytical role focused on diagnosis rather than execution, explicitly prohibiting the agent from implementing fixes or generating solutions which reduces the risk of autonomous harmful actions.
[SAFE]: The workflow incorporates a 'multi-agent-debate' requirement as an adversarial verification step, which serves to identify and correct potential reasoning flaws or biases in the diagnosis.
[PROMPT_INJECTION]: The instructions do not contain any patterns used for prompt injection, such as instructions to ignore safety filters, switch to an 'unrestricted' mode, or reveal system prompts.
[EXTERNAL_DOWNLOADS]: No external URLs for script downloads or remote code execution (e.g., curl/wget | bash) are present in the skill instructions or reference templates.
[DATA_EXFILTRATION]: The skill does not contain instructions to access sensitive system files (like credentials or private keys) or to transmit collected data to unauthorized external domains.
[COMMAND_EXECUTION]: While the skill mentions creating 'physical verification commands', these are defined as templates for user-reviewed experiments within a report and do not involve the agent executing arbitrary shell commands without oversight.

essence-diagnosis