[mirror] refactor(platform/copilot): consolidate 4 model-routing LD flags into 1 JSON flag

[yashwant86]

Mirror of upstream Significant-Gravitas#12917 for benchmark. Do not merge.

---

Summary by MergeMonkey

• Documentation:
  • Updated LD flag documentation to reflect consolidation from 4 string flags to 1 JSON flag.
  • Clarified model routing matrix and payload shape in module docstrings.
• Housekeeping:
  • Consolidated 4 separate LaunchDarkly model-routing flags into single JSON-valued flag.
  • Refactored model resolver to parse nested JSON structure instead of per-cell lookups.

[bot-mergemonkey]

⚡ Risk Assessment — NEEDS-TESTING · ~25 min review

Focus areas: JSON payload parsing and validation logic · Fallback behavior at each validation level · Test coverage for partial/invalid payloads · Backward compatibility with old flag references

Assessment: Consolidates 4 LD flags into 1 JSON flag; requires integration testing with LD service.

Changes

Files	Summary
Feature Flag Definition autogpt_platform/backend/backend/util/feature_flag.py	Replaced 4 separate string-valued model routing flags (COPILOT_FAST_STANDARD_MODEL, COPILOT_FAST_ADVANCED_MODEL, COPILOT_THINKING_STANDARD_MODEL, COPILOT_THINKING_ADVANCED_MODEL) with single JSON-valued COPILOT_MODEL_ROUTING flag supporting nested (mode, tier) structure.
Model Router Core Logic autogpt_platform/backend/backend/copilot/model_router.py	Refactored resolve_model() to fetch single JSON payload and navigate nested structure (mode → tier → model string) instead of per-cell flag lookups. Added comprehensive validation for payload shape, mode dict, tier value with granular fallback and logging at each level.
Configuration Documentation autogpt_platform/backend/backend/copilot/config.py	Updated field descriptions to reference consolidated copilot-model-routing[mode][tier] flag path instead of individual flag names.
SDK Service Documentation autogpt_platform/backend/backend/copilot/sdk/service.py	Updated docstrings and comments to reference consolidated copilot-model-routing[thinking][tier] flag path.
Test Suite Refactor autogpt_platform/backend/backend/copilot/model_router_test.py autogpt_platform/backend/backend/copilot/sdk/service_test.py	Removed _FLAG_BY_CELL import and per-cell flag tests. Added comprehensive tests for JSON payload parsing: None payload, full/partial payloads, missing modes/tiers, non-dict payloads, non-string cell values, empty strings, mode-level type errors, and single LD call regression guard.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant resolve_model
    participant LD as LaunchDarkly
    participant Logger
    participant Config
    Caller->>resolve_model: resolve_model(mode, tier, user_id)
    resolve_model->>LD: get_feature_flag_value("copilot-model-routing", user_id)
    alt LD Exception
        LD-->>resolve_model: Exception
        resolve_model->>Logger: warning("LD lookup failed")
        resolve_model->>Config: fallback
    else LD Success
        LD-->>resolve_model: payload (object)
        alt payload is None
            resolve_model->>Config: fallback
        else payload is not dict
            resolve_model->>Logger: warning("expected JSON object")
            resolve_model->>Config: fallback
        else payload is dict
            alt mode not in payload
                resolve_model->>Config: fallback
            else mode in payload but not dict
                resolve_model->>Logger: warning("mode expected dict")
                resolve_model->>Config: fallback
            else mode is dict
                alt tier value is string and non-empty
                    resolve_model-->>Caller: value.strip()
                else tier value is None
                    resolve_model->>Config: fallback
                else tier value is non-string or empty
                    resolve_model->>Logger: warning("invalid cell value")
                    resolve_model->>Config: fallback
                end
            end
        end
    end

Loading

Dig Deeper With Commands

• /review <file-path> <function-optional>
• /chat <file-path> "<question>"
• /roast <file-path>

Runs only when explicitly triggered.