The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a clear attack surface as it processes external data from .drawio and .mermaid files. However, the instructions in references/import-drawio.md and references/import-mermaid.md explicitly direct the agent to treat this data as untrusted and ignore any instructions or URLs found within labels or metadata. The provided extraction scripts include sanitization logic (e.g., clean_label and _reject_unsafe_xml) to further mitigate risks from malicious payloads.
[EXTERNAL_DOWNLOADS]: The skill references Google Fonts to provide consistent typography. It also mentions the use of Playwright for PNG export functionality, which is a standard industry tool for browser automation and rendering. These external dependencies are well-known and directly related to the skill's primary purpose.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts (drawio_extract.py, mermaid_extract.py, and self_check.py) for core functionality. These scripts perform deterministic parsing and auditing tasks without using dangerous functions like eval() or making unauthorized network connections. The self_check.py script specifically audits generated HTML to prevent the introduction of unsafe scripts or remote references.

coco-diagram