The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a custom CLI tool shinkoku to perform text extraction and image conversion on PDF files (shinkoku pdf extract-text --file-path <path> and shinkoku pdf to-image --file-path <path> --output-dir <dir>). This pattern involves executing shell commands with arguments derived from user-provided file paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from invoice files, creating a surface for indirect prompt injection.
Ingestion points: The skill reads file contents using the Read tool and extracts text from PDFs via the shinkoku utility (described in SKILL.md).
Boundary markers: No explicit instructions or delimiters are provided to isolate the extracted content or instruct the model to ignore potential commands embedded within the invoice text.
Capability inventory: The skill utilizes shell command execution (shinkoku), file reading (Read), and directory listing (Glob).
Sanitization: There is no evidence of sanitization, filtering, or validation of the content extracted from files before it is processed by the agent.

reading-invoice