The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied arguments and local repository content such as source code and logs to generate prompts that are then executed. This creates a potential surface where malicious instructions hidden within the analyzed data could influence the agent's actions.
Ingestion points: User arguments, local source files, and user-provided log or ticket data.
Boundary markers: The instructions recommend wrapping external data in tags to help distinguish it from instructions.
Capability inventory: The skill can execute shell commands and modify files once the generated prompt is approved by the user.
Sanitization: No automated sanitization or strict schema validation is specified for the data ingested from the local environment.
[EXTERNAL_DOWNLOADS]: The skill fetches technical documentation from official domains to ensure feature recommendations are current.
Evidence: Accesses documentation located at code.claude.com and platform.claude.com.
[DYNAMIC_EXECUTION]: The skill dynamically assembles a functional prompt based on its context analysis and executes that prompt immediately following user confirmation.
Evidence: The workflow concludes with a 'Run' step that executes the generated instructions as a direct agent command.

ask-claude