The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external text (dispute descriptions) which creates a surface for indirect prompt injection.
Ingestion points: scripts/dispute_analyzer.py reads user-provided text through the --input file flag or --text command-line argument.
Boundary markers: Absent; the script does not wrap input in specific delimiters or instructions to ignore embedded commands.
Capability inventory: The analysis is restricted to local regex-based extraction in scripts/dispute_analyzer.py and arithmetic in scripts/settlement_calculator.py. There are no network operations, subprocess spawns, or file system writes (outside of the user-directed --output) across any provided scripts.
Sanitization: Data is sanitized through strict regular expression filtering that only extracts recognized legal entities, dates, and monetary values, preventing the execution of embedded instructions.

mediation-analysis