The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs, Figma exports, and screenshots. This creates an attack surface where malicious instructions could be embedded in design metadata or visual content to influence the agent's behavior during the grammar generation phase.
Ingestion points: External URLs and design references are ingested via the WebFetch tool and the workflow described in SKILL.md.
Boundary markers: Absent. There are no instructions to the agent to use delimiters or to treat ingested data as non-executable text.
Capability inventory: The skill has access to Bash, Write, Edit, and WebFetch tools, allowing it to modify project files and execute commands based on the results of the analysis.
Sanitization: Not present. The workflow focuses on design measurement and rule extraction without specifying technical sanitization for content extracted from external references.
[COMMAND_EXECUTION]: The workflow instructs the agent to "execute the compiler pipeline in full" as defined in REFERENCE-COMPILER.md and to run validation commands such as /ss-score and /ss-verify. This implies the execution of shell-based scripts or tool-driven commands using the Bash tool.
[EXTERNAL_DOWNLOADS]: The skill uses WebFetch to capture and render external URLs provided as arguments or found within reference materials, connecting to remote servers to retrieve design data.

ss-reference