The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses strong directives such as 'ALWAYS activate this skill' and 'override any inclination to guess'. While these use typical injection keywords, their purpose is to enforce accuracy and prevent fabrication (anti-hallucination) rather than bypassing safety filters or security protocols.
[COMMAND_EXECUTION]: The instructions encourage the use of standard diagnostic and file-system tools (e.g., ls, cat, grep, find) and the execution of generated code to verify correctness before providing it to the user. This is consistent with the expected behavior of a development assistant skill.
[EXTERNAL_DOWNLOADS]: The skill mentions package installation (e.g., R/Python packages) but explicitly instructs the agent to 'ask first or install explicitly,' which follows safe dependency management practices.
[DATA_EXPOSURE]: The skill involves reading project files and documentation to ensure accuracy. While this accesses data from the environment, it does so to verify structure and content for the current task, with no patterns suggesting exfiltration or unauthorized access.

anti-hallucination