The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data through user prompts and web content retrieval. Ingestion points include user inputs for prompt auditing and web pages accessed via the web_fetch tool. Boundary markers are implemented through explicit operating principles in references/operating-principles.md and SKILL.md Section 0, which mandate that all external content be treated strictly as data and never as instructions. The skill's capability inventory is limited to standard search and fetch tools; it does not request file-system write access or shell execution. Sanitization is performed via instruction-based filtering enforced by the operating principles.
[SAFE]: No obfuscation, malicious command execution, or unauthorized data exfiltration patterns were detected across the skill files and its referenced principles. The skill implementation is transparent and adheres to the stated goal of professional prompt auditing.

prompt-improver