security-audit
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The methodology in SKILL.md and HUNTING.md explicitly directs the agent to build and run code from the untrusted target repository to dynamically confirm vulnerabilities. This instruction creates a path for execution of malicious code if the audited repository is adversarial.
- [DYNAMIC_EXECUTION]: Instructions guide the agent to extract suspect code segments from the target into standalone harnesses and execute them for isolation testing or fuzzing.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted codebases which can contain malicious instructions to influence the audit outcome or bypass detection. Ingestion points: Target repository source code. Boundary markers: Defined agent roles and structured reporting steps. Capability inventory: Shell subprocess execution and filesystem operations. Sanitization: Verification via independent validation agents in Phases 3 and 6.
- [COMMAND_EXECUTION]: The skill executes the included validate-findings.cjs script to verify the structural integrity of report artifacts.
- [EXTERNAL_DOWNLOADS]: The skill references configuration and methodology templates from Cloudflare's official GitHub repository, which is a trusted source.
Audit Metadata