client-instruction-schedule

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The workflow involves several command-line utilities for document processing. In SKILL.md, steps 1 and 8 specify using pdftoppm, tesseract, md5sum, soffice, unzip, and sed for OCR, file hashing, document conversion, and XML modification. These represent standard subprocess invocations within a document management context.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of processing external, untrusted litigation data.
  • Ingestion points: The skill reads an entire case folder containing diverse documents such as witness statements, correspondence, and counsel's advice (SKILL.md, Workflow Step 1).
  • Boundary markers: No specific delimiters or boundary markers are defined to isolate untrusted content from the agent's instructions during summarization.
  • Capability inventory: The skill possesses file-write capabilities (fs.writeFileSync in build-schedule-example.js) and access to multiple shell-based processing tools (sed, unzip, etc.).
  • Sanitization: There is no evidence of input validation or content sanitization applied to the extracted legal text before it is interpolated into the final Word document template.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 07:32 PM
Security Audit — agent-trust-hub — client-instruction-schedule