The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/curate_problems.py to count problems and identify coverage gaps. These commands target local project files and do not involve remote network operations.
[INDIRECT_PROMPT_INJECTION]: The skill reads external data from YAML files within the plugins/gauntlet/data/problems/ directory. This creates a surface where instructions embedded in problem content could theoretically be processed, though the use of yaml.safe_load and the requirement for human review before merging mitigate this risk. Ingestion points: plugins/gauntlet/data/problems/ directory; Boundary markers: None explicitly mentioned; Capabilities: Local file reading, shell command execution, and file writing to /tmp/; Sanitization: Utilizes yaml.safe_load for parsing data.
[DYNAMIC_EXECUTION]: The skill uses python -c to run a validation script that imports the internal gauntlet.models.BankProblem model to verify generated problem proposals. This is a low-risk operation using hardcoded validation logic.

problem-bank