The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to generate or repair agent instructions (SKILL.md) and evaluation files based on user requests. This creates a surface where untrusted user input could potentially influence the resulting logic or instructions.
Ingestion points: The workflow begins with user requests to create, scaffold, improve, or repair skills.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are specified for the generation process.
Capability inventory: The skill is intended to write and modify files in the .agents/skills/ and eng/harness-evaluation/ directories.
Sanitization: No specific validation or escaping mechanisms are mentioned for sanitizing user-provided content before it is written to the instruction files.
Local Command Execution: The instructions include a validation step using npm run lint.
This refers to executing the project's local linting suite, which is a standard part of the development and validation process for code-related artifacts.

make-skill