fix(embed): read CLI profile config as UTF-8

[koriyoshi2041]

Problem

load_config_file() opens the active profile .env with the process locale. The generated template contains UTF-8 characters, so profile startup can fail on Windows locales such as GBK or cp1252.

Fix

• read the CLI profile config explicitly as UTF-8
• add a regression that makes locale-default decoding fail while the UTF-8 path succeeds

Fixes #3837.

Test

• uv run pytest tests/test_profile_daemon_config.py -k load_config_file -q (2 passed)
• full module: 14 passed; 3 unrelated daemon-command tests require local ML dependencies and selected the documented uvx fallback
• uv run ruff check hindsight_embed/cli.py
• uv run ruff check tests/test_profile_daemon_config.py --ignore F401 (the module has three pre-existing unused imports)
• uv run ruff format --check hindsight_embed/cli.py tests/test_profile_daemon_config.py
• git diff --check

Risk

Low. The profile files are generated and written as UTF-8 already; this makes the read path match that contract. No parsing or precedence behavior changes.

[strix-security]

Strix Security Review

Warning

This pull request has 1 commit after the last Strix review (bb5f1da). Strix has not reviewed these changes.
Automatic review on push is off for this repository. To review the latest changes, tag @strix-security in a comment, or turn on re-review on push.

No security issues found.

_{Updated for bb5f1da.}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}

[Sanderhoff-alt]

The read-side change is right, but the fix looks incomplete for the locale it targets: the matching write path and the template load are still locale-dependent, so on a non-UTF-8 default locale this can turn a previously-working default profile into an unreadable one. Details inline.

[Sanderhoff-alt]

Pinning only the read side makes the default profile unreadable on a non-UTF-8 locale. hindsight-embed configure writes this exact file via CONFIG_FILE.write_text(render_config(...)) at hindsight-embed/hindsight_embed/cli.py:258 and :445, with no encoding=, so the text goes out in the process locale. render_config seeds from .env.example, which contains U+2014; both cp1252 (b'\x97') and gbk (b'\xa1\xaa') encode it happily. Before this change the write and read used the same codec and round-tripped; after it, the UTF-8 read raises UnicodeDecodeError on those bytes and startup fails. Please pin the writes too:

Suggested change

	with open(config_path, encoding="utf-8") as f:
	with open(config_path, encoding="utf-8") as f:

(plus CONFIG_FILE.write_text(render_config(config_values), encoding="utf-8") at both write sites — profile_manager.py:419 already does this).

[Sanderhoff-alt]

The description says this fixes #3837, but the same failure remains on other paths reading the same file, so a GBK/cp1252 user still hits it:

• hindsight-embed/hindsight_embed/env_template.py:24 and :28 — bundled.read_text() / candidate.read_text() with no encoding. env.example does not decode as gbk ('gbk' codec can't decode byte 0x94 in position 639), so configure and profile creation raise before any config is written.
• hindsight-embed/hindsight_embed/cli.py:1219, :1271, :1316 — config_path.read_text() with no encoding, in the config merge/set/unset commands.

Suggest pinning encoding="utf-8" on all of them in this PR so the contract holds for every reader and writer of the profile config.

[koriyoshi2041]

Good catches — I pinned UTF-8 across the template reads, both default-profile writes, and the merge/set/remove readers in a0f4ef5. I also added a regression that requires the bundled template read to request UTF-8. Focused tests pass (22 passed; 3 unrelated daemon-command tests deselected), along with Ruff, format, and diff checks.