The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires running various shell commands to install and configure the CLI tool, such as brew tap qfeius/makecli && brew install makecli and makecli integration ocr. These commands are standard for a CLI-based tool integration and point directly to the author's own package repository.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by processing untrusted data.
Ingestion points: Untrusted invoice and voucher files with extensions like .pdf, .ofd, .png, .jpg, and .jpeg are processed in SKILL.md.
Boundary markers: Present. The instructions strictly command the agent to treat only makecli integration ocr output as the recognition result and explicitly forbid using visual inspection, model knowledge, or direct HTTP calls to fill data.
Capability inventory: Local shell command execution (brew, makecli) listed in SKILL.md.
Sanitization: Absent. No text sanitization or filtering of the extracted OCR content is implemented before outputting the results.

make-integration