The Agent Skills Directory

[SAFE]: The skill serves as a guide for prompt engineering, utilizing standard libraries such as LangChain, Pydantic, and official AI SDKs. All code examples represent common development patterns and do not exhibit malicious behavior.
[SAFE]: The included optimization script (scripts/optimize-prompt.py) performs localized evaluation of prompt variations using safe concurrency patterns and does not execute remote code or access sensitive data.
[INDIRECT_PROMPT_INJECTION]: The skill involves constructing prompts that interpolate user-provided data (e.g., {query}, {text}, {problem}) into instructions for an LLM. While this is the intended purpose of the skill, it creates a surface for indirect prompt injection if user inputs are maliciously crafted. Per the primary purpose rule, this finding is considered safe.
Ingestion points: Data entering the agent context via template variables in references/details.md, references/chain-of-thought.md, and assets/prompt-template-library.md.
Boundary markers: Most examples do not include explicit boundary markers or "ignore instructions" warnings to delimit user data from system instructions.
Capability inventory: The skill facilitates calls to LLM APIs (Anthropic, OpenAI) to process these prompts and includes local execution of optimization logic in scripts/optimize-prompt.py.
Sanitization: The provided examples do not demonstrate input sanitization or filtering logic, though the skill provides separate guidance on implementing safety constraints in the "System Prompt Design" section.

prompt-engineering-patterns