Add Kimi-K2 tool simulation environment

[poofeth]

Claims https://algora.io/PrimeIntellect-ai/bounties/de4q6RZcDpDDQPXL

Summary

• adds a deterministic kimi_k2_tool_sim SingleTurnEnv for tool-call sequence simulation
• includes four provider-free scenarios covering lookup/action, refund, issue-summary, and currency/invoice workflows
• scores completions with exact tool sequence, argument matching, and final-answer rewards
• documents installation/usage and adds focused regression tests

Validation

• uv run pytest tests/test_kimi_k2_tool_sim_environment.py -q
• uv run ruff check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
• uv run ruff format --check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
• git diff --check

---

Note

Low Risk
Low risk: adds a new, self-contained example environment plus docs/tests, with no changes to shared runtime or security-sensitive logic.

Overview
Adds a new deterministic kimi-k2-tool-sim SingleTurnEnv that evaluates simulated tool-use by requiring JSON output with an exact tool-call sequence, argument matching, and final answer matching across four built-in scenarios.

Registers the environment via pyproject.toml entry point, documents install/usage and scoring, links it from environments/README.md, and adds targeted tests covering dataset construction, JSON extraction, reward partial credit, and error handling.

^{Reviewed by Cursor Bugbot for commit 169f468. Bugbot is set up for automated code reviews on this repo. Configure here.}

[poofeth]

/claim https://algora.io/PrimeIntellect-ai/bounties/de4q6RZcDpDDQPXL

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

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

^{Reviewed by Cursor Bugbot for commit 592fda7. Configure here.}

[poofeth]

Addressed both Bugbot findings in commit 169f468:

• non-dict arguments values now count the expected arguments as misses instead of removing that call from the argument denominator
• added a regression test for malformed argument maps
• added Hatch build metadata and explicit package include files to pyproject.toml

Validation:

$ uv run pytest tests/test_kimi_k2_tool_sim_environment.py -q
.......                                                                  [100%]

$ uv run ruff check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
All checks passed!

$ uv run ruff format --check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
2 files already formatted

$ git diff --check
# no output

[poofeth]

The two Bugbot inline comments appear to be stale against the original diff, but I rechecked them against the current branch:

• argument_total is incremented before malformed/non-dict actual arguments are skipped, so expected args now count as misses.
• pyproject.toml includes [build-system] and [tool.hatch.build] with Hatchling config.

Fresh validation on current HEAD:

uv run pytest tests/test_kimi_k2_tool_sim_environment.py -q
# 7 passed

uv run ruff check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
# All checks passed

uv run ruff format --check environments/kimi_k2_tool_sim tests/test_kimi_k2_tool_sim_environment.py
# 2 files already formatted

git diff --check
# no output