The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to send observational data and agent feedback to a public GitHub repository (efoo-team/skills). While this is the primary functionality, it creates a potential channel for accidental exposure of sensitive project details, credentials, or client information if the redaction guidelines are not strictly followed.
Evidence: The skill uses gh issue create and gh issue comment commands targeting the public efoo-team/skills repository.
Mitigation: The instructions include a comprehensive 'Public Repository Policy' that explicitly forbids posting credentials, client names, or private source code. It also enforces a mandatory 'Phase 6' where the user must approve a preview of the content before it is uploaded.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input provided via arguments or conversation to generate the content of GitHub issues. This input is interpolated into shell commands and issue templates.
Ingestion points: User input captured through argument-hint and the conversation flow described in Phase 1 and Phase 2 of SKILL.md.
Boundary markers: The skill uses shell heredocs with quoted delimiters (<<'EOF') and temporary files (--body-file) to ensure that user-provided text is treated as data and not as executable shell logic during the gh command execution.
Capability inventory: The skill utilizes the GitHub CLI (gh) for creating and searching issues and labels, and basic shell utilities (mktemp, cat) for temporary file management.
Sanitization: The skill mandates that the agent manually abstract and redact project-specific details (e.g., changing specific file paths to generic descriptions) before presenting the final draft to the user.

skill-memo