The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git commands (git add, git commit) to manage the version history of the prompt optimization process. These are standard developer operations intended to provide a version trail for the user's project files during the iterative workflow.
[INDIRECT_PROMPT_INJECTION]: Because the skill involves processing and classifying external text data (e.g., in Phase 1 and Phase 3), there is an inherent surface for indirect prompt injection where instructions embedded in the input data could attempt to influence the classifier. The skill mitigates this by providing templates that enforce structured JSON output and detailed label definitions with explicit boundary criteria, which limit the model's susceptibility to such inputs.
[REMOTE_CODE_EXECUTION]: The skill provides Python and R code templates in references/code-templates.md for batch processing. This code facilitates communication with official API endpoints (e.g., Anthropic) and uses established, reputable libraries such as pandas, scikit-learn, and httr2. The templates correctly handle authentication via environment variables and do not include any patterns for arbitrary remote code execution from untrusted sources.

prompt-optimizer