Fix Windows path separators and console encoding

[dc995]

Summary

Fixes three Windows-specific portability bugs discovered while running the test suite and CLI on Windows.

1. Path separators in �uild_context (product bug)

�uild_context emitted component and file-cache paths using the OS-native separator (e.g. scripts\helper.py on Windows) instead of forward slashes. This broke SARIF output and any consumer/test expecting POSIX-style paths. Fixed by normalizing relative paths with Path.as_posix().

2. Console encoding crash in the CLI (product bug)

The terminal report crashed on Windows consoles (cp1252) with UnicodeEncodeError when rendering box-drawing characters/icons. Fixed by reconfiguring stdout/stderr to UTF-8 (errors="replace") at CLI startup.

3. Test helpers (test-only)

Five test helpers built fixture file-caches with str(path), producing backslash keys on Windows. Normalized them with .as_posix() to match the product behavior.

Validation

• Unit tests: 600 passed, 11 skipped (was 11 failing on Windows before this fix)
• Lint (ruff check), format (ruff format --check), build (python -m build): all pass
• Scans across dir/single-file/zip inputs and terminal/json/markdown/sarif outputs: all work; SARIF validates against 2.1.0; exit codes correct (1 risk>50, 0 safe, 2 error)
• Integration tests (run against a real gpt-4o endpoint): the previously credential-blocked tests pass; a small number of semantic-judgment tests are nondeterministic across runs (model calibration), independent of this change

Notes

Changes are surgical: path-separator normalization and stdout encoding only. No behavior change to analysis logic.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[AbhiramDwivedi]

@keshprad — heads up: this PR has a content overlap with #87.

Both PRs fix the same str(rel) → rel.as_posix() issue across the same 6 files (build_context.py + 5 test helpers). This PR (#83) is a superset — it additionally fixes Windows UTF-8 console encoding in cli.py.

If this PR lands first, #87 becomes a full conflict on those 6 files and can be closed. Flagging so you don't have to resolve a merge conflict on review.

[rng1995]

LGTM. Clean, well-scoped cross-platform fixes:

• Normalizing relative paths with Path.as_posix() in build_context._walk_skill_files (and the matching test helpers) is the right fix — it guarantees POSIX-style component/file-cache keys regardless of OS, which keeps SARIF output and downstream pattern matching consistent on Windows.
• _ensure_utf8_streams() defensively reconfigures stdout/stderr to UTF-8 with errors="replace", guards for streams without a reconfigure method, and swallows only ValueError/OSError — a reasonable way to avoid UnicodeEncodeError on legacy Windows consoles.

Minor / optional (non-blocking):

• _ensure_utf8_streams() runs unconditionally at import on every platform, switching the error handler to replace even where the console is already UTF-8. That's harmless for a report-rendering CLI, but if you want strict behaviour preserved on POSIX you could gate it on sys.platform == "win32" or on a detected non-UTF-8 encoding.

Nice work verifying across input/output formats and confirming the previously-failing Windows tests now pass.

[rng1995]

@dc995 - Please resolve the conflicts, minor issues and merge the PR.

[dc995]

@rng1995 Done — this is ready to merge on your end:

• Conflicts resolved. They were all the str(rel) ->
  el.as_posix() lines, which fix(build_context): use forward-slash component paths (cross-platform) #87 landed independently in the meantime. I kept main's (already-merged) version, so the path-normalization is now identical to what's on main and this PR's unique remaining contribution is the cli.py UTF-8 console fix.
• Minor (non-blocking) note addressed. _ensure_utf8_streams() now skips streams that are already UTF-8, so strict encoding behaviour is preserved on POSIX and it only reconfigures where needed (e.g. legacy Windows consoles).
• Validation: ruff check + ruff format --check clean; the 120 tests across the affected modules (build_context / mcp / semantic) pass.

PR shows MERGEABLE / CLEAN / approved. I don't have merge rights on the repo, so it's over to a maintainer — please squash-merge when you're happy. (No action needed on #87; it's already merged.)

[rng1995]

@dc995 - You got another conflict to resolve. Please resolve it and I will merge the PR. Thanks!

[rng1995]

[Automated SkillSpector Review]

Re-review after rebase — re-confirming approval.

Since my approval the path-normalization (Path.as_posix()) landed independently via #87, so this PR's unique remaining source change is the cli.py UTF-8 console fix — and it now addresses my prior non-blocking nit by skipping streams that are already UTF-8 (preserving strict behavior on POSIX, reconfiguring only legacy consoles). It still guards for a missing reconfigure and swallows only ValueError/OSError. The large -403 in the diff is just uv.lock regeneration. No concerns. Approving.