vulnerability-writeup
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Interaction with Untrusted Code: The skill is designed to analyze and potentially execute Proof-of-Concept (PoC) code and build artifacts (e.g., using 'make' and './poc'). While the instructions in 'SKILL.md' emphasize the use of disposable local targets and explicit authorization, interacting with untrusted code is a core component of the skill's functionality.
- Indirect Prompt Injection Surface: This skill processes external vulnerability reports, notes, and source code, which are untrusted inputs (ingestion points identified in 'SKILL.md'). This represents a surface where data could influence agent behavior. The skill includes behavioral guardrails like 'skeptical validation' and 'source verification', but lacks explicit structural sanitization for these inputs. The agent maintains capabilities such as shell execution ('git', 'make') and file system access as noted in 'SKILL.md'.
- Command Execution Capabilities: The skill instructions involve executing shell commands such as 'git show' and various build or PoC execution commands for source verification and reproducibility checks.
- Sensitive Data Handling: The skill manages sensitive vulnerability information and includes proactive rules in 'SKILL.md' and 'references/report-format.md' against including local absolute paths in reports to prevent metadata leakage.
Audit Metadata