Appearance
项目分类
Agent QC 从分类开始。同一个仓库可以匹配多个 profiles。分类决定报告可以判断哪些风险。
按拥有的风险分类,不按语言、框架、公司或 UI 风格分类。
Profiles
| Profile | Use when the project owns | Common test focus |
|---|---|---|
agent-runtime-cli | agent loop、CLI、task execution、sandbox、tools、resume | unit、sandbox policy、protocol streams、CLI e2e、subprocess cleanup |
agent-sdk-api | public SDK、generated client、API wrappers | public signatures、fake server integration、generated contract drift |
agent-tool-mcp-gateway | tool declarations、MCP/ACP bridge、connector runtime | protocol conformance、stdio/http recovery、resource and permission refs |
multi-channel-agent-gateway | chat/channel adapters、webhooks、auth、media | channel contracts、auth/secrets、live opt-in、media routing、Docker smoke |
agent-ui-tui-desktop | GUI、TUI、desktop shell、browser-visible flows | rendering、screenshots、terminal fixtures、Playwright、accessibility |
agent-skills-plugins | skills、plugins、manifests、loaders、marketplace | schema、discovery、package boundary、fixture install、trust policy |
background-agent-scheduler | cron、queues、workers、retries、long-running agents | deterministic time、leases、checkpointing、races、stress |
agent-distribution-release | install、package、Docker、cross-platform release | package contents、install smoke、OS matrix、supply-chain scan |
agent-evals-quality | task quality、model behavior、rubrics、generated outputs | baseline comparison、semantic judge、grounding、safety/policy evals |
Mixed-profile examples
| Project shape | Profiles |
|---|---|
| Codex-like runtime with TUI and app-server protocol | agent-runtime-cli, agent-ui-tui-desktop, agent-tool-mcp-gateway, agent-sdk-api, agent-distribution-release |
| Claude Code-like local snapshot | agent-ui-tui-desktop, agent-runtime-cli, agent-sdk-api, agent-skills-plugins;metadata 缺失时 release/CI claims 标为 unknown |
| OpenClaw-like gateway and QA Lab | multi-channel-agent-gateway, agent-tool-mcp-gateway, agent-ui-tui-desktop, agent-skills-plugins, agent-distribution-release, agent-evals-quality |
| Hermes-like Python agent | agent-runtime-cli, background-agent-scheduler, agent-tool-mcp-gateway, multi-channel-agent-gateway, agent-ui-tui-desktop, agent-distribution-release |
| 带 native bridge 的 desktop GUI | agent-ui-tui-desktop, agent-tool-mcp-gateway, agent-runtime-cli, agent-skills-plugins, agent-distribution-release |
| 带 schemas/examples 的标准或文档站 | agent-distribution-release;如 schemas/CLI 被消费,也可包含 agent-sdk-api |
Classification roles
有用的 plan 会识别 owners:
| Role | Question |
|---|---|
| Profile owner | 哪个项目形态拥有风险? |
| Fact owner | 哪个系统写入被验证的事实? |
| Surface owner | 事实投影到哪个用户/运维表面? |
| Gate owner | 哪个 command、CI job、script、qcloop item 或 review 执行 gate? |
| Evidence owner | durable logs、traces、screenshots、transcripts、reports、waivers 存在哪里? |
| Risk owner | 谁决定 waiver、release 或 retry? |
Classification rules
- 按拥有的风险分类,不按语言。
- 一个仓库可以有多个 profiles;不要强行贴一个标签。
- 如果项目暴露用户可见工作,即使大部分代码是 backend/library,也要包含 surface classification。
- 如果测试需要凭证或真实 provider,标记
live-provider并显式 opt-in。 - 如果对外发布 artifact,即使是 docs-heavy project,也要包含
agent-distribution-release。 - 如果 UI 展示 runtime state,需要同时包含 surface 和 runtime/protocol gates;UI alone 不是 runtime proof。
- 如果 repo metadata 缺失,写明 limitation,不要虚构 CI/release guarantees。
- 如果 cases 重复且独立,qcloop 可以执行,但 project gates 仍需要证据。
Decision tree
text
项目是否执行 agent turns、tools、shell、sandbox 或 resume?
-> agent-runtime-cli
是否暴露 public SDK、generated client、schema 或 app-server API?
-> agent-sdk-api
是否声明、路由或桥接 tools/MCP/ACP/connectors?
-> agent-tool-mcp-gateway
是否连接 chat channels、webhooks、mobile、QR 或 media routing?
-> multi-channel-agent-gateway
用户/运维是否看到 GUI、TUI、WebUI、desktop 或 browser UI?
-> agent-ui-tui-desktop
是否加载 skills/plugins/manifests 或 marketplace assets?
-> agent-skills-plugins
是否调度 background/long-running/retry work?
-> background-agent-scheduler
是否发布 packages、Docker images、installers 或 docs site artifacts?
-> agent-distribution-release
是否用 rubrics、baselines 或 reports 判断 model/task quality?
-> agent-evals-quality1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
What classification is not
Classification 不是:
- 技术栈标签;
- 成熟度评级;
- 所有 gates 已通过的承诺;
- 独立 release checklist;
- 忽略项目 AGENTS/CONTRIBUTING 规则的理由。
Classification 只负责选择必须证明的风险和 evidence lanes。