The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides instructions to access the Linux /etc/shadow file, which contains sensitive hashed user passwords.
Evidence: unshadow /etc/passwd /etc/shadow > combined.txt in SKILL.md.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection syntax to execute shell commands automatically when the skill is loaded to check for the presence of the 'john' tool and associated wrapper scripts.
Evidence: !command -v john > /dev/null 2>&1 && echo ... and `!`test -f scripts/john/crack-linux-passwords.sh && echo ... in SKILL.md.
[COMMAND_EXECUTION]: The skill facilitates the execution of various password cracking utilities and local shell scripts.
Evidence: Commands such as john, unshadow, zip2john, rar2john, and pdf2john, as well as execution of scripts like bash scripts/john/crack-linux-passwords.sh.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied archives and hash files, creating an attack surface where malicious data could potentially influence the agent's behavior through tool outputs.
Ingestion points: Input archive files (ZIP, RAR, 7z, PDF) and hash strings processed by the various cracking and identification tools mentioned in SKILL.md.
Boundary markers: None identified in the provided instructions to delimit tool output from instructions.
Capability inventory: Reading system files, writing extracted hash files, and executing password cracking binaries.
Sanitization: No sanitization or validation of the content of the hashes or archive metadata is described.

john