codex-keysmith-instruction-installer

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The required runtime workflow ingests outsider-authored free text when the user supplies --file pointing to an external Markdown file, because the script reads that .md content and writes it into ~/.codex/*.md (which Codex CLI will later load as model instructions), i.e., the LLM context can be influenced by that outsider text.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The README explicitly instructs cloning and running code from https://github.com/Jia-Ethan/codex-keysmith.git (git clone + python3 codex-instruct.py), which fetches remote code and bundled instruction Markdown that, when executed/installed, directly controls the Codex model_instructions_file (agent prompts), so this is a runtime external dependency that can alter prompts/execute code.