The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from target codebases and API responses during its 'Analyze' and 'Exploit' phases, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Files are read using grep and cat (e.g., in references/platform-recon.md); API responses are fetched via curl (e.g., in references/exploit-techniques.md).
Boundary markers: The instructions lack explicit delimiters or warnings to treat ingested content as untrusted data.
Capability inventory: The methodology utilizes curl, grep, cat, and platform-specific tools like adb and npx to perform its tasks.
Sanitization: There is no evidence of sanitization or validation logic to filter potentially malicious instructions embedded in the target data.
[COMMAND_EXECUTION]: The skill contains numerous pre-defined shell command templates (e.g., curl, grep, adb, sqlmap, frida) for the agent to use. Although intended for security auditing, these provide a broad surface for command execution on the host or target environment.
[EXTERNAL_DOWNLOADS]: The CI/CD integration guide in references/continuous-pentest.md recommends running @keygraph/shannon via npx, which involves downloading and executing code from the NPM registry. This package is from a third-party source not included in the pre-verified trusted list.

common-pentest-methodology