The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to analyze untrusted data from external sources, creating a potential surface for indirect instructions to influence agent behavior.
Ingestion points: The methodology involves analyzing API response bodies, Swagger/OpenAPI JSON/YAML specifications, and frontend JavaScript files (e.g., app.js) as described in SKILL.md and references/api-semantic-fuzz.md.
Boundary markers: The instructions do not define clear boundaries or provide explicit directives for the agent to ignore instructions embedded within the data it processes.
Capability inventory: The skill utilizes network request tools (curl) and has the capability to save evidence and report vulnerabilities (report_vuln, evidence_save), providing a path for malicious instructions in data to trigger external actions.
Sanitization: No sanitization or validation procedures are mentioned for data ingested from remote API endpoints or documentation files.
[EXTERNAL_DOWNLOADS]: The skill's documentation references external security tools and repositories.
Evidence: Mentions the BypassPro tool located at https://github.com/0x727/BypassPro and the 403bypasser tool in references/403-bypass-patterns.md.

api-fuzz