Redesign env as incident operations workflow

Add LLM-backed grading and planning

Harden env isolation and proxy validation

Lightly optimize short prompts without mangling words

Preserve whole words and short prompts in optimizer

Add grounded optimization modes and wire dedicated optimizer module

Add PyTorch context tuning with demo-seeded retrieval and UI grounding stats

Add gold citations and multi-pack enterprise corpus switching

Add citation-aware grading and swappable corpus path

Redesign corpus and tasks around support triage and incident response

Extract UI template and reframe benchmark as enterprise workflow

Simplify UI to prompt-in optimized-prompt-out flow

Improve custom prompt retrieval relevance and project chunk ranking

Show optimized prompt output and real compression effect

Allow compressing visible chunks before selection

Use real retrieved chunks for custom prompts

Fix custom prompt UI and token optimizer flow

Initial OpenEnv submission