The Agent Skills Directory

[SAFE]: The skill content is limited to documentation, prompt templates, and evaluation rubrics. No malicious code, credential harvesting, or unauthorized actions were found.\n- [NO_CODE]: The skill does not include any executable scripts, tool definitions, or binary files. It functions as a set of instructions for the AI.\n- [EXTERNAL_DOWNLOADS]: The skill contains a reference to a GitHub repository for attribution (github.com/muratcankoylan/Agent-Skills-for-Context-Engineering). This reference is informational and does not initiate any automated network activity or code installation.\n- [PROMPT_INJECTION]: The prompt templates provided are designed for evaluation tasks and do not attempt to override system prompts or bypass safety filters.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Risk Assessment: 1. Ingestion points: The skill evaluates external AI-generated content (mentioned in 'Evaluation Task' prompt). 2. Boundary markers: The prompt templates use markdown headers to isolate the untrusted output from the evaluation instructions. 3. Capability inventory: The skill has no tools or script execution capabilities, limiting the potential impact of any injected instructions. 4. Sanitization: While no explicit sanitization is performed on input data, the lack of dangerous tools makes the risk negligible.

ai-output-evaluation