feat: add editable skill evaluation versions

backend/api/server.py

@@ -1149,6 +1149,7 @@ class EvaluationAnalyzeRequest(BaseModel):
     account_id: str = Field(min_length=1)
     query: str = Field(min_length=1, max_length=20_000)
     skill_name: str = 'label-master'
+    snapshot_id: str | None = None
     model: str | None = None
     history: list[dict[str, Any]] = Field(default_factory=list)
     context: dict[str, Any] = Field(default_factory=dict)
@@ -1162,10 +1163,19 @@ class EvaluationExperimentCreateRequest(BaseModel):
     dataset_id: str = Field(min_length=1)
     name: str = ''
     skill_name: str = 'label-master'
+    snapshot_id: str | None = None
     model: str | None = None
     concurrency: int = Field(default=2, ge=1, le=8)
 
 
+class EvaluationSkillVersionSaveRequest(BaseModel):
+    account_id: str = Field(min_length=1)
+    base_snapshot_id: str = Field(min_length=1)
+    version_name: str = Field(min_length=1, max_length=80)
+    note: str = Field(default='', max_length=1000)
+    files: dict[str, str]
+
+
 class EvaluationExperimentActionRequest(BaseModel):
     account_id: str = Field(min_length=1)
 
@@ -1534,6 +1544,74 @@ def create_app(state: AgentState) -> FastAPI:
         except EvaluationError as exc:
             raise HTTPException(status_code=400, detail=str(exc)) from exc
 
+    @app.get('/api/evaluations/skills/{skill_name}/versions')
+    async def list_evaluation_skill_versions(
+        skill_name: str,
+        account_id: str,
+    ) -> dict[str, Any]:
+        try:
+            return await asyncio.to_thread(
+                state.evaluation_runtime.list_skill_versions,
+                _safe_account_id(account_id),
+                skill_name,
+            )
+        except EvaluationError as exc:
+            raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    @app.get(
+        '/api/evaluations/skills/{skill_name}/versions/{snapshot_id}'
+    )
+    async def get_evaluation_skill_version(
+        skill_name: str,
+        snapshot_id: str,
+        account_id: str,
+    ) -> dict[str, Any]:
+        try:
+            return state.evaluation_runtime.get_skill_version_manifest(
+                _safe_account_id(account_id),
+                skill_name,
+                snapshot_id,
+            )
+        except EvaluationError as exc:
+            raise HTTPException(status_code=404, detail=str(exc)) from exc
+
+    @app.get(
+        '/api/evaluations/skills/{skill_name}/versions/{snapshot_id}/file'
+    )
+    async def get_evaluation_skill_version_file(
+        skill_name: str,
+        snapshot_id: str,
+        account_id: str,
+        path: str,
+    ) -> dict[str, Any]:
+        try:
+            return state.evaluation_runtime.read_skill_version_file(
+                _safe_account_id(account_id),
+                skill_name,
+                snapshot_id,
+                path,
+            )
+        except EvaluationError as exc:
+            raise HTTPException(status_code=404, detail=str(exc)) from exc
+
+    @app.post('/api/evaluations/skills/{skill_name}/versions')
+    async def save_evaluation_skill_version(
+        skill_name: str,
+        payload: EvaluationSkillVersionSaveRequest,
+    ) -> dict[str, Any]:
+        try:
+            return await asyncio.to_thread(
+                state.evaluation_runtime.save_skill_version,
+                _safe_account_id(payload.account_id),
+                skill_name,
+                base_snapshot_id=payload.base_snapshot_id,
+                version_name=payload.version_name,
+                note=payload.note,
+                files=payload.files,
+            )
+        except EvaluationError as exc:
+            raise HTTPException(status_code=400, detail=str(exc)) from exc
+
     @app.post('/api/evaluations/analyze')
     async def analyze_evaluation_case(
         payload: EvaluationAnalyzeRequest,
@@ -1544,6 +1622,7 @@ def create_app(state: AgentState) -> FastAPI:
                 account_id=_safe_account_id(payload.account_id),
                 query=payload.query,
                 skill_name=payload.skill_name,
+                snapshot_id=payload.snapshot_id,
                 model=payload.model,
                 history=payload.history,
                 context=payload.context,
@@ -1642,6 +1721,7 @@ def create_app(state: AgentState) -> FastAPI:
                 dataset_id=payload.dataset_id,
                 name=payload.name,
                 skill_name=payload.skill_name,
+                snapshot_id=payload.snapshot_id,
                 model=payload.model,
                 concurrency=payload.concurrency,
             )