The Agent Skills Directory

[DATA_EXFILTRATION]: The skill hardcodes absolute local filesystem paths (e.g., /Users/huimingliao/Documents/code/pentest-skills/) in its configuration and instructions. This exposes the author's local directory structure and environment details to the agent and potential users.
[CREDENTIALS_UNSAFE]: The references/example_report.md file contains hardcoded passwords and MD5 hashes (e.g., admin:123456) as part of its demonstration data. Although intended as examples, these represent the use of hardcoded secrets.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted vulnerability data from the user and interpolates it into report templates without security boundaries. \n
Ingestion points: User-provided vulnerability lists, descriptions, and proof-of-concept steps collected in the 'Collect Information' step in SKILL.md. \n
Boundary markers: No delimiters or isolation instructions are present to prevent the agent from executing instructions embedded within user-supplied vulnerability data. \n
Capability inventory: The skill's primary capability is writing files to the local reports directory. \n
Sanitization: No validation, escaping, or filtering of user-supplied text is mentioned in the workflow.

pentest-report