open-generative-ai

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill provides instructions for commands that weaken system security protections to facilitate software installation. Specifically, it guides the user to use xattr -cr to remove the macOS quarantine flag from the app bundle and sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0 to relax Linux sandbox restrictions machine-wide. These are documented as significant security changes requiring explicit user approval.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from local repositories during automated source audits.
  • Ingestion points: The scripts/audit-ogai.py script reads package.json, .gitmodules, models.js, and README.md from user-specified repository paths.
  • Boundary markers: Instructions mandate establishing "version truth" by cross-referencing multiple sources (API, package.json, source) and explicitly warn against trusting inconsistent claims in documentation files.
  • Capability inventory: The agent possesses Bash, Read, Write, Edit, and Glob tools, and executes shell commands through provided Python and Node.js scripts.
  • Sanitization: Content is processed using targeted regex and JSON parsing within a read-only Python script, avoiding the execution of untrusted input and limiting the scope of data exposure.
  • [COMMAND_EXECUTION]: The skill relies on shell commands for repository management, build processes, and auditing. The scripts/audit-ogai.py script uses subprocess.run to execute git commands (e.g., git rev-parse, git status) to verify repository integrity. The skill also provides instructions for running complex build sequences such as npm run setup and npm run electron:dev.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates downloading software from the Anil-matcha/Open-Generative-AI repository and associated submodules. It guides users through downloading platform-specific installers from GitHub and multi-gigabyte AI model weights for inference engines, while requiring confirmation of download size and target disk before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 06:29 AM
Security Audit — agent-trust-hub — open-generative-ai