The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses local configuration and artifact files within the .styleseed/ directory, as well as PRODUCT-PRINCIPLES.md and RULESETS.md. These operations are limited to the local filesystem for the purpose of design coordination and do not involve network exfiltration or access to sensitive user credentials.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files (e.g., .styleseed/bundles/<artifact-id>.md) to resolve artifact boundaries.
Ingestion points: Files located in .styleseed/artifacts/, .styleseed/bundles/, and root markdown files.
Boundary markers: None explicitly defined in the instructions.
Capability inventory: Uses Bash, Write, and Edit tools to modify STYLESEED.md based on ingested data.
Sanitization: No specific sanitization or validation of ingested file content is mentioned. While this creates a theoretical surface for indirect injection if project files are malicious, it is standard behavior for development tools and is currently assessed as safe.
[REMOTE_CODE_EXECUTION]: No patterns for remote code execution, external downloads, or piped shell commands were detected. The skill uses local tools for file manipulation and internal scoring/verification commands.

ss-restyle