test: cover brief agent-mode output

[pedropachecog]

Summary

This adds matching regression tests for rafter --agent brief scanning in Node and Python. The Node test uses spawnSync to capture both output streams, while the Python test invokes the root Typer app with color=True. Both cases require a successful exit and real Rafter content, then reject escape characters and emoji so agent mode stays plain.

Why

Issue #28 originally listed broader brief coverage, but most of that coverage landed before this change. In the issue discussion, the maintainer identified agent-mode output as the highest-priority remaining gap. Both CLIs already produce plain output; this PR adds mirrored regression coverage so a later formatting change cannot silently break that contract.

Related issue

Addresses the agent-mode portion of #28. The Node internal-helper parity work remains a separate follow-up.

Test plan

• pnpm run build
• pnpm exec vitest run tests/brief.test.ts (29 passed)
• .\.venv\Scripts\pytest.exe tests\test_brief.py -v (36 passed)
• Linux Node 20: pnpm test (1,986 passed, 53 skipped)
• Linux Python 3.11: python -m pytest tests/ -v (1,540 passed, 9 skipped)
• git diff --check upstream/main...HEAD
• node .\node\dist\index.js secrets . --diff upstream/main --engine patterns --no-auto-update --format json (zero findings)
• rafter run --mode fast (scan 77e6f82e-ebd7-422a-bc14-1f54ab73e68d): no findings in either changed test file. Those files remained byte-identical after the conflict-free rebase. The repository-wide scan also reported 20 errors, 32 warnings, and 54 notes in unmodified files.

Spec update

None. This is a test-only change and does not alter CLI behavior.

Checklist

• Both Node and Python implementations updated
• Tests pass in both (pnpm test + pytest)
• Versions match in node/package.json and python/pyproject.toml
• shared-docs/CLI_SPEC.md updated (not required because CLI behavior did not change)
• No secrets, credentials, or API keys in the diff
• AI assistance disclosed and included in the commit trailer

Prepared with OpenAI Codex. The commit includes a Co-Authored-By trailer.

[Rome-1]

Thanks — this is a clean, well-scoped addition. Checked it against the tree: the --agent flag and the Python agent-mode gating are real, the helper refactor from execFileSync to spawnSync preserves the shape the other brief tests rely on, and forcing FORCE_COLOR=1 before asserting no ANSI/emoji is exactly the right way to prove agent mode holds under a hostile env rather than passing vacuously. The Python autouse fixture that resets the global agent-mode is a nice touch.

Appreciate scoping it to the agent-mode slice and leaving the internal-helper parity as a follow-up on #28.

Merging — thanks for the contribution.