mediation-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text descriptions of disputes via its analysis script, which could potentially contain instructions aimed at influencing the agent's output.
- Ingestion points: The scripts/dispute_analyzer.py script accepts external text through the --input and --text arguments.
- Boundary markers: There are no specific delimiters or instructional guardrails implemented to prevent the agent from following prompts embedded within the dispute text.
- Capability inventory: The skill utilizes local file system access for reading inputs and writing reports, alongside script execution via the agent.
- Sanitization: The input text is processed directly using regular expressions without sanitization or filtering of instruction-like content.
- [SAFE]: The skill's functionality is transparent and confined to local operations. It relies on deterministic logic and standard Python libraries (argparse, json, re, math) for all tasks. No unauthorized network activity or attempts to access sensitive system files were found.
Audit Metadata