Fencier should be conservative with user code and secrets.

Fencier should run locally by default and should not upload source code, diffs, policies, verification results, or audit reports to a hosted service.

Secret detection should report possible secrets without printing or storing complete secret values. Audit reports should contain masked previews only.

The current scanner checks added lines from git patches and untracked files. It reports pattern id, file, optional line number, and a masked preview.

Audit reports are written to .fencier/audits/ as Markdown and JSON. Full patches are not included by default.

Setting audit.include_patch: true embeds the raw git diff in both report formats. That patch is the one part of an audit that is not masked: if the diff adds a secret, the secret appears in .fencier/audits/ in plaintext. Keep the default unless the audit directory is git-ignored and treated as sensitive local state.

Fencier verification does not prove a change is safe. It identifies risk signals and policy violations so humans and Codex can review or correct faster.

Codex integration is file and terminal based. fencier init --codex and fencier codex install write AGENTS.md; fencier codex brief reads local setup status and the latest local audit report. Neither command uploads source code, policies, audits, or findings.