Resume data is personal data. The public repository should contain sanitized examples only.

• Do not commit real candidate source files or generated private resumes.
• Use .env for secrets and local configuration.
• Validate environment variables before use.
• Treat candidate text as untrusted input for LaTeX.
• Do not send candidate data to external AI providers in the MVP.

Private candidate files should live under ignored paths such as private/ or use ignored extensions such as *.private.md. The default candidate.md, generated TeX/PDF files, and dist/ output are ignored. init uses exclusive creation and cannot overwrite an existing source. Ignore rules do not remove already tracked files.

If an LLM is added, document provider, data sent, retention/privacy tradeoffs, required environment variables, and validation gates before enabling it.

Front matter is YAML-only with a restricted schema and runtime-validated metadata; executable language markers are rejected before parsing. Candidate text is escaped for LaTeX. Compiler paths are passed as arguments without a shell. TeX uses restricted execution flags and private per-invocation working directories, which are removed afterward. PDF output is parsed and validated before publication.

The bundled compiler can download TeX support files; candidate content is not sent to an editorial service. This is a local CLI, not a sandbox service for accepting arbitrary hostile TeX files. Only compile TeX produced by Arete from validated candidate input.

Examples use fictional identities and reserved example-domain links. Sanitizing repository examples is not automatic anonymization of personal resumes, free text, Git metadata, or history.