The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data such as names for scientific terms and numerical configuration parameters. This creates an indirect prompt injection surface, which is managed through defensive programming.
Ingestion points: User-provided term names in scripts/imex_split_planner.py and various numeric parameters (dt, tolerances, orders) across all scripts via CLI arguments.
Boundary markers: No specific boundary markers or instruction-override warnings are present in the command-line examples provided in SKILL.md to wrap user input.
Capability inventory: The skill is permitted to use Read, Write, Grep, and Glob tools; the Bash tool is explicitly excluded as a security measure. Scripts write JSON-formatted results to standard output.
Sanitization: Robust sanitization is implemented in the scripts. scripts/imex_split_planner.py uses a restrictive regular expression (^[a-zA-Z_][a-zA-Z0-9_ -]*$) for term names and limits the total count and length of terms. All scripts validate numerical inputs for finiteness and positivity, and enforce maximum limits on inputs like state dimension (capped at 10 billion) and list lengths (capped at 100,000) to prevent resource exhaustion attacks.

numerical-integration