tinet-agent-cli 0.1.0.dev36__py3-none-any.whl

taco/assets/skills/README.md

@@ -0,0 +1,3 @@
+# TACO Skills
+
+运行时安装的 Agent Skill 资源位于此目录。

taco/assets/skills/taco/SKILL.md

@@ -0,0 +1,68 @@
+---
+name: taco
+description: Use when creating, configuring, debugging, publishing, or integrating TINET/Dify agents, tools, and workflows through the TACO CLI.
+---
+
+# TACO
+
+Treat the installed CLI as the source of truth. Do not rely on remembered flags, category codes, model IDs, tool IDs, or payload shapes.
+
+## Discovery protocol
+
+Before changing platform state:
+
+1. Run `taco --json profile list` and use the first row with `current=true` as the default runtime target.
+2. Run `taco help -o json`.
+3. Run `taco agent scenario list -o json`, then `taco agent scenario describe <scenario-id> -o json`.
+4. Run `taco agent schema show <resource> -o json` before editing a DSL.
+5. When the scenario names an example, run `taco agent example describe <example-id> -o json`.
+6. Run category, model, and tool discovery commands required by the scenario.
+
+Use `--json --no-input` for business commands. Parse successful data from stdout and errors from stderr. Branch on `error.code`, not translated message text.
+
+## AccessKey profile and AIKB flow
+
+TACO resolves the Zenava current profile first and falls back to its own current profile. Always use `taco --json profile list` to inspect both sources; current rows are listed first, with Zenava before TACO. To create or update a TACO-owned profile, use `taco profile set --name <profile-name> --endpoint <developer-center-endpoint> --access-key-id <id>` and provide the secret through `TACO_ACCESS_KEY_SECRET` whenever possible. `set` performs no network request. Never print the secret or use a secret-reveal option. Use `profile use` only to change TACO's current profile; pass global `--profile <name>` when an explicit target is required.
+
+For AIKB work, first run `taco aikb operation search <intent>` to obtain a small candidate set, then `taco aikb operation describe <operation-id>` before selecting and calling an unfamiliar interface. A known, previously described operation ID may be called directly. Never guess parameter names or types, bypass `--yes`, or invent an operation outside the built-in catalog. `get-file-upload-url` requires `--file` and performs the PUT upload; never expose the returned temporary upload URL. Use the returned `file_key` in a separately approved `create-file` call when needed.
+
+These rules apply equally to any AI Agent or human caller; do not branch behavior by Agent product name.
+
+## Application template flow
+
+Run `taco --json --no-input agent app template list` before choosing a template. Select only an ID returned in `recommended_apps[].app_id`; do not guess a template ID. After showing the selected template and obtaining approval, run `taco --json --no-input agent app template create <template-app-id> --yes`. Omit `--name` to 沿用模板原名称; use `--name <name>` only when the user requests an override. 创建前取得用户批准，不得自动添加 `--yes`。
+
+## Agent data OpenAPI flow
+
+For conversation history, conversation messages, or message traces, first run `taco agent operation search <intent>` to find candidates, then `taco agent operation describe <operation-id>` before using `taco agent call <operation-id>`. These developer-center APIs require an AccessKey profile; do not substitute a legacy token profile or guess parameters.
+
+Pagination is caller-controlled. When `has_more=true`, call again with the returned `next_cursor` and keep every parameter 除 cursor 外 unchanged. Stop when `has_more=false`; do not invent or parse the opaque cursor.
+
+## Safety
+
+- Never print or write AccessKey secrets, refresh tokens, access tokens, provider credentials, temporary upload URLs, or workflow secrets into a DSL or output.
+- Never guess IDs returned by platform discovery.
+- `--yes` 用于命令写操作确认；在 Agent create、Workflow create、publish、credential、secret、tool 或 install 写入前取得用户批准。
+- `--overwrite` 用于本地文件覆盖，与 `--yes` 语义不同；`agent pull --overwrite` 没有 `--yes`，但覆盖现有 DSL 前仍须单独取得用户批准。
+- Preserve local DSL on validation, debug, API, or publish failure.
+- After successful Agent publish, tell the user whether the DSL was deleted or retained.
+
+## Basic Agent flow
+
+Use scenario `agent-basic`. Discover category and model, describe the chosen model, then create or pull the Agent DSL. Read `agent schema show agent`, edit only supported fields, validate against the category, run, and publish only after approval.
+
+For continued conversations, reuse `conversation_id`. To make stop actionable, run with `--stream-events`, read the early `task_id` NDJSON event, then call `agent stop` from another process.
+
+## Builtin tool flow
+
+Use scenario `agent-builtin-tool`. Discover the exact Provider and tool, inspect its parameter and credential schemas, and configure workspace credentials only through stdin or environment variables after approval. Never copy Provider credentials or `secret-input` values into a local DSL or shell command. If a secret appears in chat or logs, treat it as compromised and request a rotated value through a secure environment variable.
+
+Only discovered `form` parameters belong in the `--parameters` file; LLM parameters are supplied at runtime and Provider credentials use credential management. Attach with `agent tool add`; do not handcraft `agent_mode.tools` or use `--replace` without separate approval. Preserve the existing `pre_prompt` while adding when to call the tool, required inputs, failure behavior, and the rule that tool results must not be invented. Validate and run both a tool-triggering query and a query that should not invoke it; confirm the SSE events show the expected tool behavior before requesting publish approval.
+
+## Workflow Tool flow
+
+Use scenario `agent-workflow-tool`. Always discover it with `agent scenario describe`, then read `agent schema show workflow` (兼容 `workflow.http`；Chatflow 用 `chatflow`)、`agent schema show agent`、`agent schema show agent-tool`，以及 named examples。Follow the guarded order `validate → dry-run → apply`: validate locally, inspect `agent workflow apply --dry-run`（Chatflow 用 `agent chatflow apply`）, request separate approval for apply, set each required secret only with `--from-env` after approval, run（Chatflow 可用 `--query`）, request Workflow/Chatflow publish approval, request Tool publish approval, then call `agent workflow tool get` for the real `workflow_tool_id` and name before attaching it to the Agent. Workflow DSL 支持 `http` / `code` / `knowledge-retrieval` / `llm` / `template-transform`，以及 `branch` / `parallel` / `iterate`；Chatflow 终点为 `answer`。
+
+Workflow/Agent debug 可能真实触发 HTTP 工具；执行前取得用户批准，并确认所有 HTTP 节点只指向专用非生产 mock/test endpoint，禁止生产客户接口。即使方法为 GET，也按可能产生外部副作用处理。
+
+Workflow must be published before Tool publish. Workflow Tool parameters are derived from Workflow Start inputs, and Agent `provider_id` is the discovered `workflow_tool_id`, not the Workflow app ID. Treat `--stream-events` lines as脱敏元数据; only the final envelope contains outputs. 不得猜测 ID，不得自动添加 `--yes`，不得手写 `agent_mode.tools`；use `agent tool add --type workflow` so the CLI resolves the current server contract.

taco/assets/skills/taco/manifest.json

@@ -0,0 +1,10 @@
+{
+  "name": "taco",
+  "format_version": 1,
+  "cli_version": "0.1.0.dev36",
+  "scenarios": [
+    "agent-basic",
+    "agent-builtin-tool",
+    "agent-workflow-tool"
+  ]
+}

taco/cli.py

@@ -0,0 +1,127 @@
+import sys
+
+import typer
+from typer._click.exceptions import Abort, ClickException, Exit
+
+from taco import __version__
+from taco.commands.agent import agent_app
+from taco.commands.aikb import aikb_app
+from taco.commands.discovery import machine_help
+from taco.commands.profile import profile_app
+from taco.core.context import CliContext
+from taco.core.errors import TacoError
+from taco.core.output import failure, to_json_line
+
+
+app = typer.Typer(help="Tinet Agent CLI Orchestrator")
+app.add_typer(agent_app, name="agent")
+app.add_typer(aikb_app, name="aikb")
+app.add_typer(profile_app, name="profile")
+app.command("help", help="以 text 或 JSON 输出当前 CLI 命令契约。")(machine_help)
+
+
+def _version_callback(value: bool) -> None:
+    if value:
+        typer.echo(__version__)
+        raise typer.Exit()
+
+
+@app.callback()
+def root(
+    ctx: typer.Context,
+    profile: str | None = typer.Option(
+        None,
+        "--profile",
+        help="使用指定 profile。",
+    ),
+    json_output: bool = typer.Option(
+        False,
+        "--json",
+        help="输出机器可读 JSON。",
+    ),
+    no_input: bool = typer.Option(
+        False,
+        "--no-input",
+        help="禁用交互输入，便于 Agent 或 CI 调用。",
+    ),
+    version: bool = typer.Option(
+        False,
+        "--version",
+        help="显示当前版本号",
+        callback=_version_callback,
+        is_eager=True,
+    ),
+) -> None:
+    ctx.obj = CliContext(
+        profile=profile,
+        json_output=json_output,
+        no_input=no_input,
+    )
+    pass
+
+
+def main() -> None:
+    try:
+        result = app(
+            args=_without_ignored_compat_args(sys.argv[1:]),
+            standalone_mode=False,
+            prog_name="taco",
+        )
+        if isinstance(result, int) and result != 0:
+            raise SystemExit(result)
+    except Exit as error:
+        raise SystemExit(error.exit_code) from None
+    except Abort:
+        _exit_main_with_error(
+            TacoError(
+                code="CLI_ABORTED",
+                message="命令已中止。",
+                exit_code=130,
+            )
+        )
+    except ClickException:
+        _exit_main_with_error(
+            TacoError(
+                code="CLI_USAGE_ERROR",
+                message="命令参数不合法。",
+                exit_code=2,
+                hint="请执行 taco help -o json 查看当前命令契约。",
+            )
+        )
+    except TacoError as error:
+        _exit_main_with_error(error)
+    except Exception:
+        _exit_main_with_error(
+            TacoError(
+                code="UNEXPECTED_ERROR",
+                message="命令执行时发生未预期错误。",
+                exit_code=1,
+                hint="请在不包含凭据的前提下提交复现步骤。",
+            )
+        )
+
+
+def _without_ignored_compat_args(args: list[str]) -> list[str]:
+    filtered: list[str] = []
+    index = 0
+    while index < len(args):
+        argument = args[index]
+        if argument == "--from-zenava":
+            index += 1
+            if index < len(args) and args[index].lower() in {"true", "false"}:
+                index += 1
+            continue
+        if argument.startswith("--from-zenava="):
+            index += 1
+            continue
+        filtered.append(argument)
+        index += 1
+    return filtered
+
+
+def _exit_main_with_error(error: TacoError) -> None:
+    if "--json" in sys.argv:
+        typer.echo(to_json_line(failure(error)), err=True)
+    else:
+        typer.echo(f"{error.code}: {error.message}", err=True)
+    raise SystemExit(error.exit_code)

taco/commands/__init__.py

@@ -0,0 +1 @@
+"""Command modules for TACO."""