The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a standard role-play prompt to define the AccessibilityAuditor specialist. No instructions to bypass safety filters, override system prompts, or ignore previous instructions were found.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths (e.g., .ssh, .env, .aws) or hardcoded credentials were detected. The skill does not perform any network operations to external or non-whitelisted domains.
[OBFUSCATION]: No obfuscated content, such as Base64 strings, zero-width characters, or homoglyphs, was found in the documentation or instructions.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill does not include any external package installations (npm, pip) or remote script execution patterns (curl | bash).
[PRIVILEGE_ESCALATION]: The skill does not use administrative commands like sudo, chmod 777, or attempts to modify system configurations.
[PERSISTENCE_MECHANISMS]: No attempts to maintain access across sessions via shell profiles, cron jobs, or registry keys were detected.
[METADATA_POISONING]: The metadata fields (name, description) accurately reflect the skill's purpose and do not contain deceptive instructions.
[INDIRECT_PROMPT_INJECTION]: While the skill is designed to process external interface code (an ingestion point), it lacks capabilities for arbitrary command execution or file writing that would make this surface critical. It remains a low-risk analytical tool.
[DYNAMIC_EXECUTION]: The skill does not generate or execute code at runtime using eval, exec, or runtime compilation.

specialist-testing-testing-accessibility-auditor