v2.0: multi-step episodes, procedural bugs, semantic grading, sessions, 71 tests

Fix the comment

Fix all things

fix: add name/difficulty to tasks, per-task [START]/[END] logs for validator

fix: stateful endpoints + score clamping for validator pass

fix: all 0.0 defaults replaced with 0.05

fix: correct [START] format, HF_TOKEN priority, remove /v1 forcing

fix: validator anchor lines and reset keyword arg

fix: use pydantic .dict() instead of dataclasses asdict, remove raise exc

fix: inline BugTriageClient to eliminate import conflict

fix: use bug_triage_client to avoid openenv-core client conflict

fix: pass task_id as positional arg to reset()

fix: /v1 base_url, verbose logging, raise on exception

remove silent fallback in call_model

use API_KEY env var from eval proxy

use injected API_BASE_URL from eval proxy

Fix all things

fix: add LOCAL_IMAGE_NAME env var to inference.py

final: inference fixed

fix: remove fallback default from HF_TOKEN

add inference.py for hackathon submission