patch-to-poc-pipeline

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches patch files, security advisories, and software artifacts from well-known and trusted domains such as GitHub, Chromium, Debian, and Ubuntu for the purpose of vulnerability analysis.
  • [COMMAND_EXECUTION]: The workflow involves orchestrating numerous security analysis and development tools, including Ghidra for reverse engineering, AFL++ for fuzzing, and Clang for compiling harnesses with sanitizers. It also uses system utilities like git, bash, and python for task automation.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes external data like patch files and source code.
  • Ingestion points: External patches (*.patch) and git repositories are read and analyzed in Phase 1 and 2.
  • Boundary markers: The methodology uses the Schema 3 structured memory format to compartmentalize analysis data and state, providing inherent separation between raw input and processing logic.
  • Capability inventory: The skill utilizes significant local capabilities including full compiler access, subprocess execution for fuzzing and analysis tools, and file system write access for PoC generation.
  • Sanitization: Vulnerability classification is performed using regex patterns on patch content to identify bug classes.
  • [PRIVILEGE_ESCALATION]: The setup instructions include the use of sudo for package management to prepare the Kali Linux research environment, which is standard for the intended use case.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM