The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code, which establishes a potential attack surface for indirect prompt injection.
Ingestion points: Source code files are read using the Read, Grep, and Glob tools as instructed in the 'Audit Checklist' and 'Scope the Audit' sections of SKILL.md.
Boundary markers: The instructions do not define specific boundary markers or delimiters for the agent to use when isolating external code content from its own instructions.
Capability inventory: The skill uses Bash and Write tools across various audit steps, which could theoretically be targeted by injected instructions from the files being analyzed.
Sanitization: No sanitization logic is specified for the input data processed during the audit.
[OBFUSCATION]: The skill contains an educational example of a homoglyph-based URL used to demonstrate security bypass techniques.
Evidence: In section 'A10: SSRF' of SKILL.md, the URL https://аllowed.com/ is used as an example of a Punycode/IDN spoof. The character 'а' is a Cyrillic homoglyph (U+0430) substituted into an otherwise Latin domain. This is provided for vulnerability identification training and is not a malicious attempt to obfuscate the skill's own operations.

owasp-audit