fix(cli): use resolved project path in doctor command

[phernandez]

Summary

• When project_root is configured (e.g., Docker), bm doctor would falsely fail because it checked for files under the original temp_path instead of the resolved path under project_root
• Now uses status.new_project.path (the actual resolved path) for all file existence checks after project creation

Fixes #665

Test plan

• just fast-check passes (lint, typecheck, impacted tests)
• just doctor passes locally (without project_root set)
• Verify with project_root configured that doctor no longer false-fails

🤖 Generated with Claude Code

[github-actions]

Claude finished @phernandez's task in 1m 44s —— View job

---

PR Review

• Read changed files and gather context
• Review code quality and standards
• Check testing coverage
• Evaluate architecture compliance
• Post final review

---

Summary

This is a correct and minimal fix. When project_root is configured (e.g., Docker deployments), the server's project_service.py ignores the user-supplied path and creates the project under {project_root}/{sanitized_name}. The doctor command was using temp_path for all file checks, which would fail since the actual files live under project_root. The fix uses status.new_project.path (the authoritative server-resolved path) instead.

Code Review Checklist

Code Quality & Standards

• Follows Basic Memory's coding conventions in CLAUDE.md
• Python 3.12+ type annotations and async patterns
• 100-character line length maintained
• No commented-out code blocks — comment explains the "why" per literate programming style
• Minimal diff — only 3 lines changed in the core logic, well-scoped
• N/A — No SQLAlchemy, FastAPI, or Typer changes in this PR

Testing & Documentation

• No test coverage for the fixed bug — There are zero tests for the doctor command in the entire codebase (tests/ and test-int/). The CLAUDE.md requires 100% coverage. Given that the doctor command is inherently an end-to-end integration scenario (real filesystem + API + sync), a basic unit test mocking project_client.create_project to return a different path would at least cover the project_path = Path(status.new_project.path) substitution. Consider as a follow-up.
• Style-only changes in the other 3 files (test_string_params_integration.py, test_tool_edit_note.py, test_coerce.py) — blank line removals, no logic changes

Basic Memory Architecture

• Correct use of status.new_project.path — consistent with how project_router.py:134 and project_management.py:370 use the same field
• No knowledge graph / markdown file integrity concerns
• Local-first principles preserved

Security & Performance

• No secrets, credentials, or sensitive data
• No new input validation surface area
• No performance concerns — read of a string field, one Path() construction

Compatibility

• Uses Path() which is cross-platform (Windows-safe)
• No emoji or unicode in console/log output

---

One Potential Follow-up: Cleanup Under project_root

doctor.py:130 — The finally block calls delete_project(project_id) without delete_notes=True:

await project_client.delete_project(project_id)

Before this fix, in project_root mode the command would fail before creating any files — so no leftover files existed. Now that the fix works correctly, doctor will successfully write files to {project_root}/doctor-{uuid}/, but the cleanup only removes the DB/config entry. The TemporaryDirectory context manager only cleans up temp_path (which is now empty), leaving test artifacts under project_root.

In a Docker deployment, this means each bm doctor run under project_root leaves behind a doctor-{uuid}/ directory. Not a blocker for merging, but worth tracking.

Suggested follow-up →

---

Overall: LGTM — The fix is correct, well-commented, and appropriately scoped. The cleanup concern above is a follow-up, not a blocker.