{"data":{"kind":"file","path":"README.md","version_id":"aiqb50i835s183kj2jrtyl6g","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1300,"modified_at":"2026-08-07T13:42:01.240000","content_hash":"b690fd7990a535d46794cf24f49a2c1f2211c538a8953a66913d4cc380fdc778"},"entries":[],"content":"# Email Assistant Environment\n\nAn RL environment that trains and evaluates language models on email-related tasks including drafting, classifying, summarizing, and replying to emails.\n\n## Tasks\n\nThe environment covers 4 core email assistant capabilities:\n\n1. **Email Drafting** - Generate professional emails from brief instructions\n2. **Email Classification** - Categorize emails by type (urgent, spam, newsletter, personal, business)\n3. **Email Summarization** - Condense long email threads into concise summaries\n4. **Email Reply** - Generate contextually appropriate replies matching tone and intent\n\n## Reward Functions\n\n| Reward | Weight | Description |\n|--------|--------|-------------|\n| `structure_reward` | 0.25 | Checks email structure (greeting, body, closing) |\n| `tone_reward` | 0.20 | Evaluates tone appropriateness (formal, casual, urgent) |\n| `completeness_reward` | 0.20 | Verifies all required elements are addressed |\n| `conciseness_reward` | 0.15 | Penalizes excessive verbosity |\n| `accuracy_reward` | 0.20 | Validates classification/summarization correctness |\n\n## Usage\n\n```python\nimport verifiers as vf\n\nenv = vf.load_environment(\"mama/email-assistant\")\n```\n\n## Evaluation\n\n```bash\nprime eval run mama/email-assistant -m Qwen/Qwen3-8B --env-args '{\"task_type\": \"draft\"}'\n```\n","encoding":"utf-8","truncated":false,"total_bytes":1300},"status":null}