F-019: docs(cli): document --allow-secrets risk in help text

[Sephyi]

Summary

docs(cli): document --allow-secrets risk in help text.

Audit context

Closes audit entry F-019 from #3.

Verification

• cargo fmt --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test --all-targets

Note: one pre-existing test porcelain_exits_within_timeout_with_no_staged_changes is a known macOS cold-start flake that reproduces on unmodified development — unrelated to this change.

[Copilot]

Pull request overview

Updates CLI help text to better communicate the security implications of --allow-secrets, addressing audit finding F-019 by warning users about potential secret exfiltration to the configured LLM provider.

Changes:

• Expanded --allow-secrets flag documentation in src/cli.rs to warn that secrets in the staged diff may be sent to the LLM.
• Added a stronger “danger” warning block in the clap help text.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This introduces a ⚠ symbol in CLI help output. The project docs note an explicit preference to replace ⚠ with a text label like WARNING: (see CHANGELOG.md / PRD.md) for better compatibility/tokenization. Consider switching to a plain-text prefix (e.g., WARNING: / DANGER:) to stay consistent and avoid terminals that render the glyph poorly.

Suggested change

	/// ⚠ DANGER: This disables the secret-scan block, meaning any API keys,
	/// DANGER: This disables the secret-scan block, meaning any API keys,

[Copilot]

The help text implies --allow-secrets simply disables the secret-scan block. In practice (see src/app.rs), secrets are still scanned and the flag only allows proceeding after an interactive confirmation; in non-interactive modes (e.g., --yes/--porcelain or piped stdin) it still fails closed with SecretsDetected. Consider updating this help text to reflect the actual behavior so users aren't surprised when --allow-secrets doesn't bypass the block in non-interactive runs.

Suggested change

	/// Allow committing even when potential secrets are detected in staged changes.
	///
	/// ⚠ DANGER: This disables the secret-scan block, meaning any API keys,
	/// credentials, or tokens present in the staged diff will be sent to the
	/// configured LLM provider as part of the prompt. Only use this flag when
	/// you have manually audited the diff and are certain no real credentials
	/// are present (e.g., test fixtures with obviously fake tokens).
	/// Allow proceeding after potential secrets are detected in staged changes.
	///
	/// ⚠ DANGER: Secrets are still scanned for. This flag does not bypass
	/// detection; it only allows an interactive confirmation to continue after
	/// secrets are found. In non-interactive modes (for example --yes,
	/// --porcelain, or piped stdin), commitbee still fails closed instead of
	/// proceeding. If you do confirm, any API keys, credentials, or tokens
	/// present in the staged diff will be sent to the configured LLM provider
	/// as part of the prompt. Only use this flag when you have manually
	/// audited the diff and are certain no real credentials are present
	/// (e.g., test fixtures with obviously fake tokens).