Spaces:
Sleeping
Sleeping
Kandukuri
feat: OpenEnv spec compliance - LLM agent, /state endpoint, pyproject.toml, uv.lock, .gitignore
b621598 | name: repair-strategy-system-v2 | |
| version: "2.0.0" | |
| description: > | |
| A deterministic environment that simulates a distributed system with hidden state. | |
| Agents must infer root causes from metrics and ambiguous logs. | |
| Enforces cascading failures, delayed recovery, and action costs. | |
| environment: | |
| module: env_v2 | |
| class: RepairEnvV2 | |
| seed: 42 | |
| tasks: | |
| - id: easy | |
| description: "Database failure with clear symptoms. Single root cause." | |
| max_steps: 10 | |
| optimal_steps: 3 | |
| - id: medium | |
| description: "Cache failure and passive queue degradation. Misleading symptoms." | |
| max_steps: 15 | |
| optimal_steps: 5 | |
| - id: hard | |
| description: "Full cascading failure starting from DB. API and Queue both impacted." | |
| max_steps: 20 | |
| optimal_steps: 7 | |
| - id: multi_root | |
| description: "Independent failures in both DB and Queue. High complexity." | |
| max_steps: 25 | |
| optimal_steps: 10 | |
| action_space: | |
| type: structured | |
| model: models_v2.Action | |
| fields: | |
| action_type: | |
| type: enum | |
| values: | |
| - restart_service | |
| - scale_up | |
| - clear_cache | |
| - repair_database | |
| - inspect_database | |
| - inspect_api | |
| - inspect_cache | |
| - inspect_queue | |
| - no_op | |
| target: | |
| type: enum | |
| values: | |
| - database | |
| - api | |
| - cache | |
| - queue | |
| observation_space: | |
| type: structured | |
| model: models_v2.SymptomObservation | |
| fields: | |
| metrics: | |
| type: dict | |
| description: "Map of component name to its metrics (latency, error_rate, etc.)" | |
| logs: | |
| type: list | |
| description: "List of highly ambiguous log entries from various components" | |
| step_count: | |
| type: int | |
| max_steps: | |
| type: int | |
| inspections_remaining: | |
| type: int | |
| description: "Counter for free inspections (first 3 are free)" | |
| reward: | |
| model: models_v2.Reward | |
| range: [-5.0, 10.0] | |
| description: > | |
| Dense reward: +1.5 for root cause repair started, +0.5 for symptom improvement, | |
| -1.0 for wrong component targeted, +5.0 terminal bonus, -0.2 step penalty. | |
| Includes Action Costs (V3): e.g., -0.5 for DB/Scale, -0.2 for Restart/Cache. | |
| graders: | |
| module: graders_v2 | |
| function: grade_v2 | |
| baseline: | |
| module: baseline_v2 | |
| function: run_task | |
| server: | |
| host: "0.0.0.0" | |
| port: 7860 | |
| endpoints: | |
| - path: /tasks | |
| method: GET | |
| - path: /grader | |
| method: POST | |
| - path: /baseline | |
| method: GET | |
| - path: /diagnose | |
| method: POST | |
| description: "Submit a theoretical diagnosis for immediate feedback" | |
| - path: /reset | |
| method: POST | |
| - path: /step | |
| method: POST | |
| - path: /state | |
| method: GET | |