The Agent Skills Directory

Command Execution: The sub-agents managed by this skill are instructed to perform technical verification by running build tools and executing scripts found within the microsoft/winappcli repository. This capability is used to empirically verify the feasibility of design proposals. The skill includes instructions to ensure these operations are confined to temporary directories and do not modify the main repository tree.
Indirect Prompt Injection Surface: The skill ingests user-provided specifications which could potentially contain instructions intended to influence the agent's behavior. 1. Ingestion points: SKILL.md captures the specification text from user-provided file paths or descriptions. 2. Boundary markers: The skill does not use specific structural delimiters but relies on explicit instructions for sub-agents to ignore the specification as its own evidence. 3. Capability inventory: Sub-agents can read files (grep, glob, view) and execute repository scripts and SDK tools (makeappx, signtool, build-cli.ps1). 4. Sanitization: The skill relies on the Team Lead Test and mandatory independent research to sanitize the impact of potentially malicious inputs.
Dynamic Execution: Sub-agents are encouraged to perform 'cheap experiments,' such as building throwaway projects in temporary directories. This allows the agent to verify how tools and APIs actually behave, rather than relying solely on documentation or user claims. This dynamic execution is scoped to technical verification within the repository's context.

spec-review