security-audit-workflow

Pass

Audited by Gen Agent Trust Hub on Oct 31, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The tools/run.py script uses subprocess.run() to execute Docker commands and native binaries. These commands are built using a list of arguments, which is a secure practice to mitigate shell injection risks.
  • [DYNAMIC_EXECUTION]: The tools/tools.json configuration for govulncheck, cargo-audit, and bandit includes shell commands that dynamically install packages from official registries (Go, Rust, and Python) into Docker containers at runtime before execution.
  • [PRIVILEGE_ESCALATION]: The skill facilitates mounting the host's Docker socket (/var/run/docker.sock) for the trivy-image and dockle tools. As noted in the skill's tools/README.md, this grants the container full control over the host's Docker daemon.
  • [INDIRECT_PROMPT_INJECTION]: The audit workflow ingests and processes untrusted data from the target repository, such as source code and documentation (PRDs, TechSpecs, OpenAPI contracts). This creates an attack surface where malicious instructions embedded in project files could potentially influence the agent's behavior.
  • Ingestion points: Project source files, docs/prd.md, docs/techspec.md, and api-contract.yaml.
  • Boundary markers: The skill does not explicitly define delimiters or instructions for the agents to ignore embedded prompts in the analyzed content.
  • Capability inventory: The agent can execute shell commands via the Docker wrapper, read/write files, and access the network.
  • Sanitization: There is no explicit sanitization or filtering of external project content described in the workflow.
  • [EXTERNAL_DOWNLOADS]: The skill downloads official Docker images and vulnerability databases from well-known and trusted services, including Docker Hub, PyPI, Crates.io, and the Go module proxy.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 31, 2026, 01:05 PM