The Agent Skills Directory

[SAFE]: The skill implements a standard Eval-Driven Development (EDD) framework. It provides instructions and examples for defining and running evaluation tests on code changes within a development environment. It follows security best practices by recommending human review for security-sensitive tasks.\n- [COMMAND_EXECUTION]: The skill examples demonstrate the use of standard local development tools such as grep, npm test, and npm run build. These operations are used for code validation and regression testing within the local project context and do not involve remote access or unauthorized system modifications.\n- [INDIRECT_PROMPT_INJECTION]: The skill establishes an evaluation workflow that ingests data from external sources. (1) Ingestion points: Project source code (e.g., src/auth.ts) and session outputs from AI models. (2) Boundary markers: Uses markdown headers and specific labels such as [CAPABILITY EVAL]. (3) Capability inventory: Employs Read, Write, Edit, Bash, Grep, and Glob tools as defined in SKILL.md. (4) Sanitization: No explicit sanitization or filtering is defined for the ingested content. The identified surface is considered safe and intended for development testing purposes.

eval-harness