rlm-pattern

Pass

Audited by Gen Agent Trust Hub on Mar 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses system tools such as unzip, tar, sed, awk, and wc for file decompression and chunking.
  • [COMMAND_EXECUTION]: It generates and executes a shell script to partition CSV data, which involves dynamic command construction based on file line and column counts.
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it reads and processes large, untrusted data files and transmits findings between multiple agent layers.
  • Ingestion points: Reads local files (source code, logs, CSVs) for analysis.
  • Boundary markers: Employs semantic boundaries and line offsets to divide content.
  • Capability inventory: Executes shell commands and has the ability to write to temporary directories.
  • Sanitization: No explicit logic is provided for sanitizing or escaping the data extracted from files before it is analyzed.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 18, 2026, 10:12 AM
Security Audit — agent-trust-hub — rlm-pattern