The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of evaluation logic that processes untrusted data from application traces and datasets. This creates a surface for indirect prompt injection where malicious content could influence scoring. Evidence: 1. Ingestion points: traces and datasets in Phase 2. 2. Boundary markers: uses interpolation templates like {{ inputs }} in Phase 7 instructions without mandated delimiters. 3. Capability inventory: Bash, Write, Read, Grep, and Glob tools used for code generation. 4. Sanitization: no explicit sanitization or escaping instructions provided for the generated prompts.
[DYNAMIC_EXECUTION]: The skill generates runnable Python code for the user and uses the Bash tool to execute an introspection script for environment discovery. This runtime code generation is consistent with the skill's purpose as an evaluation prototype generator.
[COMMAND_EXECUTION]: The skill uses the Bash tool to run a Python-based introspection script that identifies available MLflow scorers in the local environment, ensuring recommendations are accurate based on the installed software version.

build-a-scorer