The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes text content from user-provided PDF files to generate revision reports, creating a surface where malicious instructions embedded in a document could influence the agent.
Ingestion points: External PDF files defined in references/craft-steps.json are loaded via the doc_open tool.
Boundary markers: SKILL.md contains warnings regarding content truncation and data types (images vs text), but does not provide explicit delimiters or "ignore instructions" guards for the ingested content.
Capability inventory: The skill has file-write capabilities through the doc_compare_report tool and command execution capabilities for local scripts.
Sanitization: No evidence of text sanitization or escaping is present for the content extracted during the comparison process.
[COMMAND_EXECUTION]: The instructions require the agent to execute local system commands, specifically calling a Python script (scripts/write_script.py) and the pdfcraft-cli tool to perform its primary functions.
[DYNAMIC_EXECUTION]: The skill uses a local Python script to dynamically generate a JSON-formatted execution script (steps.json) based on a template. This generated file is then passed to the CLI tool for execution, representing a runtime orchestration pattern.

pdfcraft-revision-compare