The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository's source code by searching for specific comment markers, creating a surface for indirect prompt injection.
Ingestion points: Reads repository files via the grep tool (defined in SKILL.md) to extract content following the ponytail: marker.
Boundary markers: Employs a regex filter to isolate comments, but lacks explicit instructions to ignore potentially malicious instructions embedded within those comments.
Capability inventory: Includes file system reading via grep, metadata retrieval via git blame, and the ability to write reports to the file system.
Sanitization: No evidence of input sanitization or escaping is present in the instructions.
[COMMAND_EXECUTION]: The skill relies on local shell commands to perform its auditing functions.
Evidence: Uses grep -rnE --exclude-dir=.git --exclude-dir=node_modules --exclude-dir=dist --exclude-dir=build '(#|//|/[*]) ?ponytail:' . to scan the project.
Evidence: Uses git blame -L<line>,<line> to identify the author of specific lines of code.

ponytail-debt