feat: add real code execution — run_code, run_tests, submit_fix tools

feat: convert to MCP tool-calling pattern matching reference repos

fix: shorten HF short_description to under 60 chars

feat: major upgrade — multi-step MDP, AST injection, 5-signal reward

fix: send reward=None on reset, clamp all scores to strict (0,1)

fix: clamp all scores strictly to (0, 1) — no 0.0 or 1.0

fix: use only platform-injected API_BASE_URL and API_KEY for LiteLLM proxy

fix: use API_BASE_URL and API_KEY directly for LiteLLM proxy

fix: prioritize HF_TOKEN over API_KEY for LiteLLM proxy compliance

fix: clamp all task scores to strict (0,1) range — never exactly 0.0 or 1.0

fix: prioritize API_KEY over HF_TOKEN for LiteLLM proxy compliance

docs: update baseline scores to latest measured results (composite 0.50 → 0.69)

fix: pass task param to env.reset(), improve medium/hard prompts

fix: guarantee [START]/[END] stdout output even on env connection failure

fix: add pyrightconfig, update pyproject packages, polish inference & models

Add uv.lock for multi-mode deployment (required by openenv validate)

Fix: use LOCAL_IMAGE_NAME (matches submission form)

KW-WM proof: reward prediction beats baselines by 38.8-71.7%

Match official sample inference.py: update API_BASE_URL default to HF router

Fix critical crash bug + audit fixes

Fix all 4 remaining issues for hackathon

Update baselines with real post-fix LLM scores

Fix 4 critical grader bugs

Update baseline with real GPT-4o-mini scores

Rename to Knowledge-Work World Model (KW-WM) + add citations

Real-world utility boost: MBRL research pipeline

Improve grader quality + research framing

CRITICAL: Redirect [DEBUG]/[SUMMARY] to stderr

Fix env var format to match submission checklist exactly

Fix inference.py log format to match official spec

Add .env to gitignore

CodeReviewEnv v1.0 — OpenEnv-compliant submission