The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python test scripts (scripts/iconpark_tool_test.py and scripts/xml_lint_test.py) that use subprocess.run to execute CLI commands. These are utilized solely for internal testing and validation of the skill's own tooling.
[DYNAMIC_EXECUTION]: The skill employs dynamic execution for compatibility and testing infrastructure.
scripts/xml_text_overlap_lint_test.py uses os.execv to perform process replacement, acting as a compatibility wrapper to execute the xml_lint_test.py script.
scripts/xml_lint_test.py contains a compile() call, which is used for internal evaluation of test logic or XML fragments during validation suites.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and editing existing slide content, creating a surface for indirect prompt injection if those slides contain adversarial instructions.
Ingestion points: Untrusted data enters the agent context through the outputs of lark-cli slides +xml-get and xml_presentation.slide.get as described in SKILL.md.
Boundary markers: While data is processed as structured XML, there are no specific markers or instructions to the model to ignore natural language commands embedded within slide text elements.
Capability inventory: The skill can execute shell commands via the lark-cli binary and perform local file system writes for XML drafts and screenshots.
Sanitization: All generated or modified content is passed through an automated layout and schema validator (scripts/xml_lint.py) which ensures structural integrity and visual compliance.

lark-slides