exploitability-validation

Pass

Audited by Gen Agent Trust Hub on Jun 7, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is a sophisticated security tool designed for validating vulnerability reports. It demonstrates a high level of rigor through its structured pipeline and multiple validation gates (e.g., GATE-1 through GATE-8).
  • [COMMAND_EXECUTION]: The skill facilitates the execution of various internal utility scripts located in the libexec/ directory (e.g., mantishack-validation-helper, mantishack-run-feasibility) and compiles proof-of-concept (PoC) code using compilers like gcc. Crucially, it mandates that all such operations must be performed via libexec/mantishack-run-sandboxed, which restricts network access, limits resource usage, and constrains file system writes. This is an appropriate security control for a tool of this nature.
  • [PROMPT_INJECTION]: While the skill contains strong instructional language (e.g., "CRITICAL", "MUST", "GATES"), these are used to define a strict methodology for the agent to follow and do not represent malicious injection attempts or behavior overrides. The skill also incorporates an "Indirect Prompt Injection" attack surface (Category 8) because it analyzes external source code and findings.
  • Ingestion points: Vulnerability findings in findings.json and source code from the target repository.
  • Boundary markers: The skill uses a structured multi-stage pipeline (Stages A-F) and JSON-based work documents to separate reasoning and data.
  • Capability inventory: File writing (via tools), compilation and execution of binaries (sandboxed), and script execution (sandboxed) in SKILL.md, stage-a-oneshot.md, and stage-e-feasibility.md.
  • Sanitization: The skill relies on its sandboxed execution environment to prevent malicious code in the analyzed repository from impacting the host system.
  • [EXTERNAL_DOWNLOADS]: The skill mentions a soft dependency on the z3-solver Python package for its SMT-backed analysis features. This is a well-known, legitimate academic and security library.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 7, 2026, 01:29 PM