Add SWE-Swiss SFT environment

[poofeth]

Summary

• add a lightweight SWE-Swiss SFT train/eval environment backed by public Hugging Face SWE-Swiss datasets
• normalize messages rows into Verifiers prompt/question/answer/info examples, supporting both JSON-string and list message formats
• score completions with normalized text similarity against the reference assistant response

Bounty: https://algora.io/PrimeIntellect-ai/bounties/FHr4fCjVPjFDmFED

Submitted as the custom Algora bounty claim for the SWE-Swiss training-data environment lane.

Scope note: this targets the existing SWE-Swiss training-data environment path, not the full auto-generation pipeline tier. It keeps runtime requirements light and does not require a sandboxed repository checkout.

Validation

• uv run pytest tests/test_swe_swiss_sft_environment.py -q
• uv run ruff check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
• uv run ruff format --check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
• git diff --check
• live smoke-loaded one row from SWE-Swiss/SWESwiss-SFT-Repair-4K into prompt, question, answer, and info columns

---

Note

Low Risk
Low risk: adds a new standalone environment package plus tests and docs, without modifying core verifier/runtime logic. Main risk is dataset-shape assumptions (messages formatting) affecting usability for some SWE-Swiss dataset variants.

Overview
Adds a new installable swe-swiss-sft SingleTurn environment backed by Hugging Face SWE-Swiss SFT datasets, converting each row’s messages transcript into prompt/question/answer examples (supports list or JSON-string messages).

Introduces a lightweight reward (SequenceMatcher-based normalized text similarity) and exposes configurable dataset/split/example-limit/shuffle arguments via load_environment, with accompanying environment README, packaging (pyproject.toml entry point), a brief mention in environments/README.md, and new unit tests covering parsing, dataset loading, reward behavior, and lazy environment construction.

^{Reviewed by Cursor Bugbot for commit 16e512b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[poofeth]

Validation evidence for the bounty review:

$ uv run pytest tests/test_swe_swiss_sft_environment.py -q
.....                                                                    [100%]

$ uv run ruff check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
All checks passed!

$ uv run ruff format --check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
2 files already formatted

$ git diff --check
# no output

$ uv run python - <<'PY'
# loaded one example through load_swe_swiss_dataset(num_examples=1, shuffle_seed=None)
# result: 1 ['prompt', 'question', 'answer', 'info']
PY

This PR targets the existing-data path for https://algora.io/PrimeIntellect-ai/bounties/FHr4fCjVPjFDmFED.

[poofeth]

Proactive follow-up in commit a2dc196:

• listed swe_swiss_sft in environments/README.md

Validation after the update:

$ uv run pytest tests/test_swe_swiss_sft_environment.py -q
.....                                                                    [100%]

$ uv run ruff check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
All checks passed!

$ uv run ruff format --check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
2 files already formatted

$ git diff --check
# no output

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a2dc196. Configure here.}

[poofeth]

Addressed Bugbot feedback in commit 16e512b:

• coalesce null role/content values before stringifying, so content: null is skipped instead of becoming the literal string "None"
• added regression coverage for null-content rows

Validation:

$ uv run pytest tests/test_swe_swiss_sft_environment.py -q
......                                                                   [100%]

$ uv run ruff check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
All checks passed!

$ uv run ruff format --check environments/swe_swiss_sft tests/test_swe_swiss_sft_environment.py
2 files already formatted

$ git diff --check
# no output

[poofeth]

/claim https://algora.io/PrimeIntellect-ai/bounties/FHr4fCjVPjFDmFED