fix(search): keep grep fallback root searchable

[wanquanY]

What does this PR do?

Fixes the search_files grep fallback silently returning zero matches for its default path="." search.

The fallback intentionally passes --exclude-dir='.*' to match ripgrep's hidden-directory behavior. grep also applies that glob to the command-line search root, however, so a relative root such as . (or ./directory) excludes itself before traversal begins.

This change anchors relative grep roots at the terminal backend's live, shell-quoted $PWD. It preserves the existing hidden-directory exclusion, avoids resolving remote/container paths on the host, and keeps the user-provided path component shell-escaped.

Related Issue

N/A — searched open and closed issues/PRs for the error and --exclude-dir behavior; no duplicate found.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Anchor relative roots passed by ShellFileOperations._search_with_grep() at the backend's live $PWD.
• Replace false-green handcrafted grep tests with production-API tests that force the real grep fallback.
• Cover the default ., an explicit ./directory, the .hub prompt-injection cache, and an arbitrary hidden directory.

How to Test

1. Run python -m pytest tests/tools/test_search_hidden_dirs.py tests/tools/test_search_error_guard.py tests/tools/test_file_operations.py -q (80 passed, 2 skipped on macOS 26.2).
2. Run the search-focused cross-suite selection (39 passed, 113 deselected).
3. Run python -m ruff check tools/file_operations.py tests/tools/test_search_hidden_dirs.py and git diff --check.
4. On Debian 12 with GNU grep, the old relative-root command returns exit 1 with zero lines; the anchored-root command returns the visible match and still excludes both named and arbitrary dot-directories.

The canonical full macOS run reached 29,484 passed / 306 skipped, with 22 unrelated platform/environment failures and one isolated test-file timeout; every touched and search-related suite passed. The PR's Ubuntu CI is the authoritative full-suite gate.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass (see macOS full-run note above; scoped suites are green)
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 26.2 (BSD grep) and Debian 12 arm64 container (GNU grep)

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A; behavior is internal and covered by an inline rationale
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide
• I've updated tool descriptions/schemas if I changed tool behavior — N/A; public behavior is restored, not changed

Screenshots / Logs

# Regression mutation (old implementation)
1 failed: expected a visible match, got total_count == 0

# Fixed implementation
80 passed, 2 skipped

# Debian 12 / GNU grep
old_exit=1 old_lines=0 new_exit=0 new_lines=1
/work/my-skill/SKILL.md:2:This is a real skill.

[alt-glitch]

This was generated by AI during triage.

Related: #48878 and #34017 address the same grep fallback family via conditional hidden-root handling. This PR instead anchors relative roots at the backend PWD while preserving hidden-descendant exclusion.

[kshitijk4poor]

Salvaged via #84780 — your commit was cherry-picked onto current main with your authorship preserved, and merges as soon as checks complete (auto-merge armed, rebase).

Review notes from the salvage: your "$PWD" anchoring was selected over the competing #81205 (cd && pwd -P subprocess approach) — zero extra exec per search, resolution happens in the sandbox shell itself, no rg result-shape change, and the production-API test rewrite is a strict fidelity upgrade over the old handcrafted-grep tests. Verified with a live E2E matrix (./sub/./sub/absolute/space+quote/injection-probe roots) plus a mutation check (reverting the prod file makes your new tests fail).

Thanks for the fix!