The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON files and incorporates their content (such as metric names) into its output summary. This creates a surface where an attacker could embed malicious instructions in the input files to influence the agent's behavior during the balancing workflow.
Ingestion points: The scripts/validate_balance_runs.py script reads files passed to the --input and --spec arguments.
Boundary markers: There are no explicit delimiters or instructions to the agent to disregard potential commands embedded within the input data.
Capability inventory: The skill executes a local script to perform comparisons; the agent context may have broader file-system and tool access.
Sanitization: The script performs type validation for numerical fields but does not sanitize or escape string fields that are echoed in the final JSON output.

auto-balancer