fix(beta/filesystem): pin read/write encoding to UTF-8

[genisis0x]

Summary

FilesystemToolkit exposes read_file, write_file, and update_file tools that the model uses to manipulate user files on disk. All three called Path.read_text() / Path.write_text() without an explicit encoding, so each operation inherited locale.getpreferredencoding() — cp1252 on most Windows installs:

• read_file raised UnicodeDecodeError on any file whose bytes weren't valid cp1252 (most modern source files, JSON / YAML with non-ASCII strings, Markdown with smart quotes, anything with emoji or CJK glyphs).
• write_file raised UnicodeEncodeError on any payload the model generated that wasn't representable in cp1252 (the same set of characters).
• update_file did the read and the write in the same call against the platform default, so a round-trip on Windows silently transcoded the file every time the model edited it.

Same class of bug as #1731 / #2814 / #2815, this time on the LLM-facing beta toolkit.

Fix

Pass encoding=\"utf-8\" explicitly at all three call sites so the toolkit produces the same bytes on Linux, macOS, and Windows. No change to the public tool signatures; no new options. Inline comments explain why the kwarg is required so a future refactor doesn't drop it again.

Test

test/beta/tools/test_filesystem.py gets three regression tests that drive the toolkit through Agent.ask (per test/beta/CLAUDE.md's public-API guidance) and assert byte-level equality against the explicit UTF-8 encoding of the payload — not against Path.read_text(), which would mask the bug on POSIX:

• test_write_file_encodes_as_utf8
• test_read_file_decodes_as_utf8
• test_update_file_preserves_utf8 — replaces an ASCII token with a CJK glyph so the read+write pair has to round-trip non-cp1252 content.

npm test equivalent: .venv/bin/python -m pytest test/beta/tools/test_filesystem.py reports 16 passed.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
autogen/beta/tools/toolkits/filesystem.py	96.62% <100.00%> (ø)

... and 62 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[marklysze]

Clean fix — pinning encoding="utf-8" on all three file-touching operations (read_file, write_file, update_file) is the correct solution for the Windows cp1252 default. The three new tests cover the non-ASCII roundtrip contract exactly right: write the bytes directly, read back and compare at the byte level. LGTM.

[marklysze]

Correct and consistent with the rest of the series (#2819, #2825, #2826, #2827, #2814, #2815).

#2818 — FilesystemToolkit read/write/update encoding

Three-method fix — _read_file, _write_file, _update_file — all pinned to UTF-8. The update_file change is the most important: without matching encoding on both the read and write side, a Windows machine iterating on a UTF-8 source file would silently re-encode output as cp1252 each edit cycle.

Tests pin the contract at the byte level — write non-ASCII payload, read back raw bytes and compare against payload.encode("utf-8"). Same proven pattern as the other PRs in this series.

One note: the inline comments in the production code are more verbose than necessary (the PR body already carries this context), but they don't affect correctness.