The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to process and execute existing prompt artifacts, which introduces a surface for indirect prompt injection.
Ingestion points: Processes promptFiles and requirements provided via user input or workspace context as described in SKILL.md.
Boundary markers: No explicit boundary markers or 'ignore instructions' delimiters are defined within the provided skill files.
Capability inventory: The skill utilizes Prompt Tester and Prompt Evaluator to execute prompts and evaluate results within a sandbox environment.
Sanitization: No explicit sanitization or filtering of the input prompt content is performed before processing.
Context: The use of a deterministic sandbox (.copilot-tracking/sandbox/) provides isolation for these operations, which is a standard practice for handling potentially untrusted prompt content during refactoring tasks.

prompt-refactor