maxc-cli 0.5.0__tar.gz → 0.6.0__tar.gz

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: maxc-cli
-Version: 0.5.0
+Version: 0.6.0
 Summary: Agent-native MaxCompute CLI for external coding agents
 Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3.9
@@ -31,7 +31,7 @@ MaxCompute 元数据、SQL、作业、权限和数据传输等数据面操作。
 ### 公共云（推荐）
 
 ```bash
-# Alibaba Cloud CLI 需要 >= 3.3.3；旧版本先运行 aliyun upgrade
+# aliyun maxc 需要 Alibaba Cloud CLI >= 3.3.19
 aliyun version
 
 # OAuth 是交互式登录的首选方式，不需要把长期 AK/SK 放进命令行
@@ -49,6 +49,9 @@ aliyun maxc query cost "SELECT * FROM schema.table WHERE ds='20260415'" --json
 aliyun maxc query "SELECT * FROM schema.table WHERE ds='20260415'" --json
 ```
 
+支持自升级的非 Homebrew Alibaba Cloud CLI 3.3.5+ 可在用户确认后运行
+`aliyun upgrade`；更早版本、Homebrew 安装或缺失 CLI 请按官方安装方式更新。
+
 如果 Alibaba Cloud CLI 扩展不可用，或明确需要 PyPI 版本，可使用独立入口。
 独立入口要求 Python 3.9 或更高版本：
 
@@ -110,13 +113,15 @@ Agent 启动时生成一次 32 位小写十六进制 session ID，并在整个
 UA="AlibabaCloud-Agent-Skills/alibabacloud-maxcompute-cli/<session-id>"
 ```
 
-每条 `aliyun maxc` 命令都追加 `--user-agent "$UA"`。然后依次运行：
+每条调用云 API 的 `aliyun maxc` 命令都追加 `--user-agent "$UA"`；本地
+help、`agent context`、`agent manifest`、`session show` 和 `cache status`
+可以省略。然后依次运行：
 
 ```bash
-aliyun maxc agent context --user-agent "$UA" --json          # 仅检查本地版本、配置和能力；不访问网络
-aliyun maxc agent manifest --user-agent "$UA" --json         # 从实时 parser 生成命令、参数和副作用清单
+aliyun maxc agent context --json                              # 仅检查本地版本、配置和能力；不访问网络
+aliyun maxc agent manifest --json                             # 从实时 parser 生成命令、参数和副作用清单
 aliyun maxc agent doctor --online --user-agent "$UA" --json  # 验证身份与后端可达性
-aliyun maxc agent skill --user-agent "$UA" --json            # Skill 路径、名称与 min_cli_version
+aliyun maxc agent skill --json                                # Skill 路径、名称与 min_cli_version
 ```
 
 只有 `agent doctor --online` 能证明远端已就绪；不要把 `agent context` 中的
@@ -139,10 +144,19 @@ aliyun maxc agent skill --user-agent "$UA" --json            # Skill 路径、
       {
         "id": "meta.search",
         "title": "Search tables",
-        "command": "maxc meta search <keyword> --json",
+        "command": "maxc --user-agent <user_agent> meta search <keyword> --json",
         "executable": false,
-        "placeholders": {"keyword": "search keyword"},
-        "args_schema": {},
+        "placeholders": {
+          "keyword": "<keyword>",
+          "user_agent": "<user_agent>"
+        },
+        "args_schema": {
+          "user_agent": {
+            "type": "string",
+            "description": "Reuse the User-Agent generated once for the current Agent session.",
+            "pattern": "^AlibabaCloud-Agent-Skills/alibabacloud-maxcompute-cli/[0-9a-f]{32}$"
+          }
+        },
         "effect": "read",
         "confirmation_required": false,
         "agent_allowed": true
@@ -178,7 +192,7 @@ maxc meta describe my_table --json
   "mode": "read_only",
   "force": false,
   "allowed_operations": ["SELECT"],
-  "effective_hints": {"odps.sql.read.only": "true"},
+  "effective_hints": {},
   "policy_decision": "allowed"
 }
 ```
@@ -219,10 +233,14 @@ src/maxc_cli/
 
 ## 限制
 
-- **查询安全**：公共 Agent Skill 的 SQL 契约只支持 `SELECT`，不通过
-  `query` 或 `job submit` 执行 DDL/DML。`data upload`、
-  `data download --overwrite` 和 `job cancel` 是独立的有副作用操作，仍需
-  与影响相匹配的明确授权。
+- **查询安全**：SQL 默认按只读请求处理。只有用户明确要求具体 DDL/DML
+  时，才核对完整 statement、project、schema、目标和影响，通过 `query`
+  或 `job submit` 一次提交一条语句并显式增加 `--force`。`data upload`、
+  `data download --overwrite` 和 `job cancel` 是独立的有副作用操作，同样需
+  与影响相匹配的明确授权。权限、账号、project、system、resource、package
+  等管理 SQL 和未知语法不在公共 `--force` 正向允许列表中。前置 `SET`
+  同样属于执行上下文：项目安全、访问控制和脱敏参数始终被阻断，强制写入
+  仅接受已审查的语句级执行参数。
 - **OAuth 优先**：公共云交互式登录优先 OAuth。只有运行环境明确要求时才用
   AK/SK、STS、环境变量或外部凭证进程。直接 AK/SK 会写入
   `~/.maxc/config.yaml`（文件权限 0600）。OAuth 需要账号/组织已分配官方
@@ -251,3 +269,45 @@ src/maxc_cli/
 pip install -e .
 pytest tests/ -m unit
 ```
+
+## Trusted egress authentication
+
+In a sandbox whose trusted egress injects ODPS credentials, explicitly use:
+
+```bash
+aliyun maxc auth login-proxy --project <project> --odps-endpoint <https-endpoint> --json
+aliyun maxc agent doctor --online --json
+```
+
+This saves `auth.provider: proxy` and routing information only. MaxC does not
+load AK/SK, STS, OAuth or credential helpers, and sends no Authorization or ODPS
+token headers. The egress must authenticate the sandbox and inject credentials.
+TLS verification remains enabled; preserve the egress-provided CA environment.
+Only proceed when `agent doctor --online` reports `data.ready=true`. A saved
+configuration (`--no-validate`) alone does not prove connectivity or identity.
+The egress must cover metadata, SQL polling and result/Tunnel endpoints actually
+used by the workload; ordinary endpoints without injection will reject requests.
+To return to local signing, explicitly run the appropriate `auth login` command.
+
+## Versioned semantic packages
+
+`maxc semantic` manages remote Catalog packages using the current connection.
+Specify the verified main-account ID with `--namespace`. Available operations:
+`list`, `get`, `revisions`, `export`, `create`, `diff`, `apply`, `publish`, `delete`.
+`meta semantic` continues to manage local annotations.
+
+Export a USER_DRAFT, edit its `object`/`content`, review `diff`, then use its
+`planDigest` with `apply --plan-digest ... --yes`. Publishing is a separate
+`publish --expected-spec-id ... --expected-revision ... --yes` operation.
+Run each command's `--help` for required arguments.
+
+The editable JSON format is `maxc.semantic/v1`; its identity and revision fields
+bind edits to one account, endpoint and object. Whole sections replace prior
+content; omitted sections remain unchanged and explicit empty arrays clear them.
+Exports strip server-generated table IDs and preserve unknown nested fields.
+Source files and output paths must not traverse symlinks/reparse points.
+
+Metadata edits lack server CAS; publication lacks idempotency keys. An uncertain
+write must be reconciled before retrying. Publication checks are not SQL or
+business validation. History lists the latest 20 summaries, and an immutable
+revision read does not guarantee historical DataBridge analysis.

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/README.md

@@ -9,7 +9,7 @@ MaxCompute 元数据、SQL、作业、权限和数据传输等数据面操作。
 ### 公共云（推荐）
 
 ```bash
-# Alibaba Cloud CLI 需要 >= 3.3.3；旧版本先运行 aliyun upgrade
+# aliyun maxc 需要 Alibaba Cloud CLI >= 3.3.19
 aliyun version
 
 # OAuth 是交互式登录的首选方式，不需要把长期 AK/SK 放进命令行
@@ -27,6 +27,9 @@ aliyun maxc query cost "SELECT * FROM schema.table WHERE ds='20260415'" --json
 aliyun maxc query "SELECT * FROM schema.table WHERE ds='20260415'" --json
 ```
 
+支持自升级的非 Homebrew Alibaba Cloud CLI 3.3.5+ 可在用户确认后运行
+`aliyun upgrade`；更早版本、Homebrew 安装或缺失 CLI 请按官方安装方式更新。
+
 如果 Alibaba Cloud CLI 扩展不可用，或明确需要 PyPI 版本，可使用独立入口。
 独立入口要求 Python 3.9 或更高版本：
 
@@ -88,13 +91,15 @@ Agent 启动时生成一次 32 位小写十六进制 session ID，并在整个
 UA="AlibabaCloud-Agent-Skills/alibabacloud-maxcompute-cli/<session-id>"
 ```
 
-每条 `aliyun maxc` 命令都追加 `--user-agent "$UA"`。然后依次运行：
+每条调用云 API 的 `aliyun maxc` 命令都追加 `--user-agent "$UA"`；本地
+help、`agent context`、`agent manifest`、`session show` 和 `cache status`
+可以省略。然后依次运行：
 
 ```bash
-aliyun maxc agent context --user-agent "$UA" --json          # 仅检查本地版本、配置和能力；不访问网络
-aliyun maxc agent manifest --user-agent "$UA" --json         # 从实时 parser 生成命令、参数和副作用清单
+aliyun maxc agent context --json                              # 仅检查本地版本、配置和能力；不访问网络
+aliyun maxc agent manifest --json                             # 从实时 parser 生成命令、参数和副作用清单
 aliyun maxc agent doctor --online --user-agent "$UA" --json  # 验证身份与后端可达性
-aliyun maxc agent skill --user-agent "$UA" --json            # Skill 路径、名称与 min_cli_version
+aliyun maxc agent skill --json                                # Skill 路径、名称与 min_cli_version
 ```
 
 只有 `agent doctor --online` 能证明远端已就绪；不要把 `agent context` 中的
@@ -117,10 +122,19 @@ aliyun maxc agent skill --user-agent "$UA" --json            # Skill 路径、
       {
         "id": "meta.search",
         "title": "Search tables",
-        "command": "maxc meta search <keyword> --json",
+        "command": "maxc --user-agent <user_agent> meta search <keyword> --json",
         "executable": false,
-        "placeholders": {"keyword": "search keyword"},
-        "args_schema": {},
+        "placeholders": {
+          "keyword": "<keyword>",
+          "user_agent": "<user_agent>"
+        },
+        "args_schema": {
+          "user_agent": {
+            "type": "string",
+            "description": "Reuse the User-Agent generated once for the current Agent session.",
+            "pattern": "^AlibabaCloud-Agent-Skills/alibabacloud-maxcompute-cli/[0-9a-f]{32}$"
+          }
+        },
         "effect": "read",
         "confirmation_required": false,
         "agent_allowed": true
@@ -156,7 +170,7 @@ maxc meta describe my_table --json
   "mode": "read_only",
   "force": false,
   "allowed_operations": ["SELECT"],
-  "effective_hints": {"odps.sql.read.only": "true"},
+  "effective_hints": {},
   "policy_decision": "allowed"
 }
 ```
@@ -197,10 +211,14 @@ src/maxc_cli/
 
 ## 限制
 
-- **查询安全**：公共 Agent Skill 的 SQL 契约只支持 `SELECT`，不通过
-  `query` 或 `job submit` 执行 DDL/DML。`data upload`、
-  `data download --overwrite` 和 `job cancel` 是独立的有副作用操作，仍需
-  与影响相匹配的明确授权。
+- **查询安全**：SQL 默认按只读请求处理。只有用户明确要求具体 DDL/DML
+  时，才核对完整 statement、project、schema、目标和影响，通过 `query`
+  或 `job submit` 一次提交一条语句并显式增加 `--force`。`data upload`、
+  `data download --overwrite` 和 `job cancel` 是独立的有副作用操作，同样需
+  与影响相匹配的明确授权。权限、账号、project、system、resource、package
+  等管理 SQL 和未知语法不在公共 `--force` 正向允许列表中。前置 `SET`
+  同样属于执行上下文：项目安全、访问控制和脱敏参数始终被阻断，强制写入
+  仅接受已审查的语句级执行参数。
 - **OAuth 优先**：公共云交互式登录优先 OAuth。只有运行环境明确要求时才用
   AK/SK、STS、环境变量或外部凭证进程。直接 AK/SK 会写入
   `~/.maxc/config.yaml`（文件权限 0600）。OAuth 需要账号/组织已分配官方
@@ -229,3 +247,45 @@ src/maxc_cli/
 pip install -e .
 pytest tests/ -m unit
 ```
+
+## Trusted egress authentication
+
+In a sandbox whose trusted egress injects ODPS credentials, explicitly use:
+
+```bash
+aliyun maxc auth login-proxy --project <project> --odps-endpoint <https-endpoint> --json
+aliyun maxc agent doctor --online --json
+```
+
+This saves `auth.provider: proxy` and routing information only. MaxC does not
+load AK/SK, STS, OAuth or credential helpers, and sends no Authorization or ODPS
+token headers. The egress must authenticate the sandbox and inject credentials.
+TLS verification remains enabled; preserve the egress-provided CA environment.
+Only proceed when `agent doctor --online` reports `data.ready=true`. A saved
+configuration (`--no-validate`) alone does not prove connectivity or identity.
+The egress must cover metadata, SQL polling and result/Tunnel endpoints actually
+used by the workload; ordinary endpoints without injection will reject requests.
+To return to local signing, explicitly run the appropriate `auth login` command.
+
+## Versioned semantic packages
+
+`maxc semantic` manages remote Catalog packages using the current connection.
+Specify the verified main-account ID with `--namespace`. Available operations:
+`list`, `get`, `revisions`, `export`, `create`, `diff`, `apply`, `publish`, `delete`.
+`meta semantic` continues to manage local annotations.
+
+Export a USER_DRAFT, edit its `object`/`content`, review `diff`, then use its
+`planDigest` with `apply --plan-digest ... --yes`. Publishing is a separate
+`publish --expected-spec-id ... --expected-revision ... --yes` operation.
+Run each command's `--help` for required arguments.
+
+The editable JSON format is `maxc.semantic/v1`; its identity and revision fields
+bind edits to one account, endpoint and object. Whole sections replace prior
+content; omitted sections remain unchanged and explicit empty arrays clear them.
+Exports strip server-generated table IDs and preserve unknown nested fields.
+Source files and output paths must not traverse symlinks/reparse points.
+
+Metadata edits lack server CAS; publication lacks idempotency keys. An uncertain
+write must be reconciled before retrying. Publication checks are not SQL or
+business validation. History lists the latest 20 summaries, and an immutable
+revision read does not guarantee historical DataBridge analysis.

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/__init__.py

@@ -2,4 +2,4 @@
 
 __all__ = ["__version__"]
 
-__version__ = "0.5.0"
+__version__ = "0.6.0"

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/_samples.py

@@ -12,6 +12,15 @@ SAMPLES: dict[str, str] = {
         'maxc query "SELECT 1"\n'
         "maxc meta list-tables --project my_proj"
     ),
+    "semantic.list": "maxc semantic list --namespace <account_id> --json",
+    "semantic.get": "maxc semantic get sales --namespace <account_id> --source PUBLISHED --revision <revision> --json",
+    "semantic.revisions": "maxc semantic revisions sales --namespace <account_id> --json",
+    "semantic.export": "maxc semantic export sales --namespace <account_id> --output sales.json --json",
+    "semantic.create": "maxc semantic create sales --namespace <account_id> --file definition.json --json",
+    "semantic.diff": "maxc semantic diff sales --namespace <account_id> --file sales.json --json",
+    "semantic.apply": "maxc semantic apply sales --namespace <account_id> --file sales.json --plan-digest <reviewed_digest> --json",
+    "semantic.publish": "maxc semantic publish sales --namespace <account_id> --expected-spec-id <spec_id> --expected-revision <reviewed_revision> --json",
+    "semantic.delete": "maxc semantic delete sales --namespace <account_id> --expected-spec-id <spec_id> --json",
     # ── query ──────────────────────────────────────────────────────────────
     "query": (
         'maxc query "SELECT 1"\n'
@@ -24,6 +33,9 @@ SAMPLES: dict[str, str] = {
         "maxc auth login --oauth\n"
         "maxc auth login --from-env --json"
     ),
+    "auth.login-proxy": (
+        "maxc auth login-proxy --project <project> --odps-endpoint <endpoint> --json",
+    ),
     "auth.login-external": (
         "maxc auth login-external --process-command 'credential-helper --format json'"
     ),

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/app.py

@@ -939,7 +939,11 @@ class MaxCApp:
                     status="pending",
                     data={
                         "job_id": job.job_id,
-                        "safety": build_safety_block(force=force, sql=sql),
+                        "safety": build_safety_block(
+                            force=force,
+                            sql=sql,
+                            effective_hints=job.effective_hints,
+                        ),
                     },
                     metadata={
                         "job_id": job.job_id,
@@ -985,7 +989,11 @@ class MaxCApp:
                     status="pending",
                     data={
                         "job_id": job.job_id,
-                        "safety": build_safety_block(force=force, sql=sql),
+                        "safety": build_safety_block(
+                            force=force,
+                            sql=sql,
+                            effective_hints=job.effective_hints,
+                        ),
                     },
                     metadata={
                         "job_id": job.job_id,
@@ -1108,6 +1116,7 @@ class MaxCApp:
                 self.log(command, envelope.status, envelope.metadata)
                 return envelope
             result.extra_metadata.update(execution_metadata)
+            result.effective_hints = dict(job.effective_hints)
             for warning in common_warnings:
                 if warning not in result.warnings:
                     result.warnings.append(warning)
@@ -1285,7 +1294,14 @@ class MaxCApp:
         envelope = Envelope(
             command="job.submit",
             status="pending",
-            data={"job_id": job.job_id, "safety": build_safety_block(force=force, sql=sql)},
+            data={
+                "job_id": job.job_id,
+                "safety": build_safety_block(
+                    force=force,
+                    sql=sql,
+                    effective_hints=job.effective_hints,
+                ),
+            },
             metadata={
                 "job_id": job.job_id,
                 "project": job.project,
@@ -2761,6 +2777,19 @@ class MaxCApp:
             raise ValidationError("--schema must not be empty.")
         return target_project, target_schema
 
+    def semantic_manage(self, operation: str, *, namespace: str, **options) -> 'Envelope':
+        """Run a remote SemanticSpec workflow, separate from local annotations."""
+        from .semantic_management import SemanticManager
+
+        manager = SemanticManager(self.backend, namespace)
+        data = manager.execute(operation, **options)
+        command = "semantic." + operation
+        metadata = {"namespace": namespace}
+        envelope = Envelope(command=command, status="success", data=data,
+                            metadata=metadata, agent_hints=AgentHints(warnings=manager.warnings))
+        self.log(command, envelope.status, metadata)
+        return envelope
+
     def semantic_set(
         self,
         table_name: 'str',
@@ -4864,6 +4893,37 @@ class MaxCApp:
             envelope.metadata["continuation_expires_at_unix"] = expires_at
         return envelope
 
+    def auth_login_proxy(
+        self, *, project: str, endpoint: str,
+        region_name: "str | None" = None, tunnel_endpoint: "str | None" = None,
+        no_validate: bool = False, target_config_path: "Path | None" = None,
+    ) -> Envelope:
+        """Save an explicit, credential-free egress authentication configuration."""
+        target_path = target_config_path or default_global_config_path()
+        auth = AuthConfig(provider="proxy", project=project, endpoint=endpoint,
+                          region_name=region_name, tunnel_endpoint=tunnel_endpoint)
+        resolved = resolve_auth_connection(self.config, auth_override=auth)
+        if no_validate:
+            identity = {"authenticated": None, "configured": True,
+                        "validation_status": "configuration_only", "auth_type": "proxy",
+                        "identity_source": "egress_proxy", "project": resolved.project,
+                        "endpoint": resolved.endpoint}
+            warnings = ["Configuration saved without remote validation; run agent doctor --online."]
+        else:
+            identity, warnings = self._validate_auth_config(auth)
+        # A failed identity check must leave the previous login untouched.
+        migrate_legacy_session_override(target_path)
+        persist_login_config(target_path, auth=auth)
+        envelope = Envelope(
+            command="auth.login-proxy", status="success",
+            data={"identity": identity,
+                  "persistence": {"saved": True, "validated": not no_validate}},
+            metadata={"config_path": str(target_path)},
+            agent_hints=AgentHints(actions=[action("auth.whoami")], warnings=warnings),
+        )
+        self.log("auth.login-proxy", envelope.status, envelope.metadata)
+        return envelope
+
     def auth_login_external(
         self,
         *,
@@ -5599,6 +5659,8 @@ class MaxCApp:
             has_creds = bool(effective_settings.get("external_process_command"))
         elif provider == "oauth":
             has_creds = auth_cfg.oauth.is_configured()
+        elif provider == "proxy":
+            has_creds = True
         else:
             has_creds = False
 
@@ -5756,13 +5818,8 @@ class MaxCApp:
         failed_check_ids = [
             check["id"] for check in checks if check["status"] == "fail"
         ]
-        command = current_cli_entry_point()
         if ready and not online:
-            actions = [SuggestedAction(
-                id="agent.doctor.online",
-                title="Verify credentials and backend",
-                command=f"{command} agent doctor --online --json",
-            )]
+            actions = [action("agent.doctor", data={"online": True})]
         elif ready:
             actions = [action("meta.list-tables", data=context_data)]
         elif "auth.configured" in failed_check_ids:
@@ -6557,20 +6614,31 @@ class MaxCApp:
 
         # Build actions
         qe_actions: list[SuggestedAction] = []
+        write_operations = known_write_operations(result.sql_executed)
         if result.tables_used and result.extra_metadata.get("result_kind") != "statement":
             qe_actions.append(action("meta.describe", data=data, metadata=metadata))
         if result.has_more and next_cursor:
             qe_actions.append(action("query.paginate", data=data, metadata=metadata))
         elif result.has_more and result.job_id:
             qe_actions.append(action("job.result", data=data, metadata=metadata))
-        if dry_run:
+        if dry_run and not write_operations:
             qe_actions.append(action("job.submit", data=data, metadata=metadata))
+        elif dry_run and write_operations:
+            warnings.append(
+                "No replay action was generated for mutating SQL. A dry-run does not "
+                "authorize later execution; rerun the exact statement with `--force` "
+                "only after the user confirms that mutation."
+            )
 
         # Add safety block
         if command in {"job.wait", "job.result"}:
             data["safety"] = build_observational_safety_block(command)
         else:
-            data["safety"] = build_safety_block(force=force, sql=result.sql_executed)
+            data["safety"] = build_safety_block(
+                force=force,
+                sql=result.sql_executed,
+                effective_hints=result.effective_hints,
+            )
 
         return Envelope(
             command=command,
@@ -6592,6 +6660,8 @@ class MaxCApp:
         analysis: 'dict[str, Any]',
         force: 'bool' = False,
     ) -> 'Envelope':
+        analysis = dict(analysis)
+        effective_hints = analysis.pop("_effective_hints", {})
         warnings = list(analysis.get("warnings", []))
         insights = []
         if analysis.get("estimated_input_size_bytes") == 0:
@@ -6606,14 +6676,26 @@ class MaxCApp:
 
         # Build actions
         ae_actions: list[SuggestedAction] = []
-        if command == "query.cost":
-            ae_actions.append(action("query.explain", data=analysis, metadata=metadata))
-        ae_actions.append(action("query", data=analysis, metadata=metadata))
+        write_operations = known_write_operations(sql)
+        if not write_operations:
+            if command == "query.cost":
+                ae_actions.append(action("query.explain", data=analysis, metadata=metadata))
+            ae_actions.append(action("query", data=analysis, metadata=metadata))
+        else:
+            warnings.append(
+                "No replay action was generated for mutating SQL. Cost or plan analysis "
+                "does not authorize later execution; rerun the exact statement with "
+                "`--force` only after the user confirms that mutation."
+            )
         if analysis.get("tables_used"):
             ae_actions.append(action("meta.describe", data=analysis, metadata=metadata))
 
         # Add safety block
-        analysis["safety"] = build_safety_block(force=force, sql=sql)
+        analysis["safety"] = build_safety_block(
+            force=force,
+            sql=sql,
+            effective_hints=effective_hints,
+        )
 
         return Envelope(
             command=command,

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/audit.py

@@ -18,6 +18,7 @@ _SENSITIVE_KEY_FRAGMENTS = (
     "continuation",
     "cookie",
     "credential",
+    "logview",
     "password",
     "process_command",
     "refresh_token",

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/auth_providers.py

@@ -137,6 +137,8 @@ class ResolvedAuthConnection:
             pyodps_catalog.Client instance, or None if the SDK is not
             installed or catalog endpoint is unavailable.
         """
+        if self.provider == "proxy":
+            return None
         try:
             from maxcompute_tea_openapi import models as open_api_models
             from pyodps_catalog.client import Client as CatalogClient
@@ -171,6 +173,8 @@ def auth_settings_available(config: 'MaxCConfig') -> 'bool':
     settings, _, _suppressed = resolve_odps_settings(config)
     provider = infer_auth_provider(config, settings)
     try:
+        if provider == "proxy":
+            return not missing_odps_settings(settings, auth_type="proxy")
         if provider == "sts_token":
             return not missing_odps_settings(settings, auth_type="sts_token")
         if provider == "external":
@@ -193,6 +197,32 @@ def resolve_auth_connection(
     settings, sources, suppressed_env_vars = resolve_odps_settings(config, auth_override=auth_override)
     provider = infer_auth_provider(config, settings, auth_override=auth_override)
 
+    if provider == "proxy":
+        missing = missing_odps_settings(settings, auth_type="proxy")
+        if missing:
+            raise ValidationError(
+                f"Proxy authentication is missing required fields: {', '.join(missing)}.",
+                suggestion="Run `auth login-proxy --project <project> --odps-endpoint <endpoint>`.",
+            )
+        from .proxy_auth import ProxyAccount
+
+        # Never retain stale credentials or helper configuration in this mode.
+        connection_fields = {"provider", "project", "endpoint", "region_name",
+                             "tunnel_endpoint", "catalog_endpoint"}
+        settings = {key: value if key in connection_fields else None
+                    for key, value in settings.items()}
+        return ResolvedAuthConnection(
+            auth_type="proxy", provider="proxy",
+            project=settings["project"], endpoint=settings["endpoint"],
+            region_name=settings.get("region_name"),
+            tunnel_endpoint=settings.get("tunnel_endpoint"),
+            catalog_endpoint=settings.get("catalog_endpoint"),
+            access_id=None, secret_access_key=None, security_token=None,
+            token_expires_at=None, identity_source="egress_proxy",
+            settings=settings, setting_sources=sources,
+            suppressed_env_vars=suppressed_env_vars, account=ProxyAccount(),
+        )
+
     if provider == "external":
         missing = missing_odps_settings(settings, auth_type="external")
         if missing:
@@ -329,7 +359,7 @@ def infer_auth_provider(
 ) -> 'str':
     auth = auth_override or config.auth
     explicit = (auth.provider or settings.get("provider") or "").strip().lower()
-    if explicit in {"access_key", "sts_token", "sts", "external", "oauth"}:
+    if explicit in {"access_key", "sts_token", "sts", "external", "oauth", "proxy"}:
         if explicit == "sts":
             return "sts_token"
         return explicit

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/backend/data.py

@@ -326,8 +326,10 @@ class DataMixin:
             raise ValidationError(
                 f"`{definition.name}` is a view; views are read-only and cannot be loaded via Tunnel.",
                 suggestion=(
-                    "Choose a physical table that supports Tunnel upload, or use "
-                    "an approved data-write workflow outside the public Agent Skill."
+                    "Choose a physical table that supports Tunnel upload. If the "
+                    "user separately authorizes one exact supported SQL write and "
+                    "its target and effect, submit that statement with --force; "
+                    "otherwise use an approved bulk data-write workflow."
                 ),
             )
         partition_columns = {c.name for c in definition.partition_columns}
@@ -352,8 +354,9 @@ class DataMixin:
             raise ValidationError(
                 f"Columns {unsupported} have complex types not supported by CSV upload.",
                 suggestion=(
-                    "Convert the input through an approved data-write workflow "
-                    "outside the public Agent Skill."
+                    "Use a format and approved data-write workflow that supports "
+                    "the exact complex types. Do not auto-convert or submit a SQL "
+                    "write unless the user separately authorizes its target and effect."
                 ),
             )
 

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/backend/job.py

@@ -781,8 +781,10 @@ class JobMixin(QueryMixin):
         return sql.rstrip(";") if sql else None
 
     def _safe_logview(self, instance) -> 'str | None':
-        """Safely get logview URL from instance."""
+        """Get a LogView URL without exposing its signed query or fragment."""
+        from ..utils import sanitize_logview_url
+
         try:
-            return instance.get_logview_address()
+            return sanitize_logview_url(instance.get_logview_address())
         except Exception:
             return None

{maxc_cli-0.5.0 → maxc_cli-0.6.0}/src/maxc_cli/backend/odps.py

@@ -20,6 +20,7 @@ from .data import DataMixin
 from .job import JobMixin
 from .meta import MetaMixin
 from .query import _write_operation
+from .semantic import SemanticMixin
 
 # When the pyodps instance tunnel is unavailable, pyodps emits a UserWarning
 # and falls back to the CSV result reader, which caps results at 10000 rows.
@@ -58,6 +59,7 @@ def _summarize_fallback_warning(message: 'str') -> 'str | None':
 class OdpsBackend(
     JobMixin,  # JobMixin extends QueryMixin
     CatalogMixin,
+    SemanticMixin,
     MetaMixin,
     DataMixin,
     AuthMixin,