The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data including source code and Storybook files, creating a potential surface for indirect prompt injection.\n
Ingestion points: The scripts audit-drift.sh, extract-component-usage.sh, and audit-archived.sh read project source files (.tsx, .stories.tsx) to extract metadata and usage tallies.\n
Boundary markers: SKILL.md includes explicit instructions for the agent to follow a 'Log-only rule' for out-of-scope issues and requires user confirmation before committing findings.\n
Capability inventory: The skill utilizes Bash, Read, and Write capabilities to execute local scripts and generate JSON reports within the .storybook/ directory.\n
Sanitization: Input processing relies on regular expressions and basic string parsing; there is no formal isolation or execution of the content being analyzed.\n- [DYNAMIC_EXECUTION]: Several scripts utilize dynamic loading or execution methods for core functionality.\n
Evidence: extract-component-usage.sh executes an embedded Python script via a heredoc, and refresh-usage.sh dynamically locates and invokes other scripts located within the vendor bundle's directory hierarchy. Additionally, SKILL.md includes examples for executing local Node.js scripts for app-graph extraction.

sb-audit