The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents provided by users, which can be used to inject malicious instructions into the agent's context.
Ingestion points: The skill reads SOP sources in multiple formats, including PDF, xlsx, docx, and images, as specified in SKILL.md and references/source_format_handling.md.
Boundary markers: There are no explicit instructions for the agent to use delimiters or to ignore potential instructions embedded within the source documents.
Capability inventory: The skill allows writing files to the local system (SKILL.md) and provides code for executing shell commands and Python scripts (references/source_format_handling.md).
Sanitization: The instructions do not mention any validation or sanitization of the content extracted from the source files before it is processed or written to the output JSON.
[COMMAND_EXECUTION]: The skill provides the agent with specific commands and script patterns to execute for document parsing.
Evidence: references/source_format_handling.md contains instructions for using CLI tools like pdftotext and tesseract, and Python libraries such as pypdf, openpyxl, and pytesseract to extract data from various file formats.
[CREDENTIALS_UNSAFE]: The skill is designed to process and output configuration data that may contain embedded secrets.
Evidence: The Gotchas section in SKILL.md explicitly mentions handling URLs that include secrets (e.g., https://USER:TOKEN@host/...) and instructs the agent to embed them verbatim in the output JSON file.

automation-sandbox-post-copy-config-generate