no-slop

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

No Slop

No Slop is a cold, precise anti-slop workflow. Its job is to detect generic AI UI, remove it, and block weak output before it ships.

After anti-slop remediation, hand app-like surfaces to

reality-skill

so the result is not only less generic, but also usable beyond a static mockup.

No Slop是一套冷静、精准的反敷衍工作流，其职责是检测通用AI生成的UI，将其移除，并在交付前拦截低质量输出。

完成反敷衍整改后，将类应用界面交付给

reality-skill

，确保最终成果不仅不再千篇一律，还能超越静态原型具备实际可用性。

Command Interface

命令接口

Use the command grammar in

references/command-interface.md

Primary commands:

bash

no-slop --scan [target]
no-slop --fix [target]
no-slop --redesign [target]
no-slop --judge [target]
no-slop --prevent [target-or-brief]
no-slop --autopsy [target-or-brief]
no-slop -e --scan [target]

Short flags:

```
-s
```
,
```
--scan
```
: scan and score.
```
-f
```
,
```
--fix
```
: surgical remediation.
```
-r
```
,
```
--redesign
```
: full redesign.
```
-j
```
,
```
--judge
```
: run the strict judge panel.
```
--prevent
```
: prevention mode before UI generation.
```
--autopsy
```
: emit the forensic AI UI Autopsy report.
```
-e
```
,
```
--economy
```
: disable judge subagents and use local deterministic checks only.

Default mode is

--scan

. Judge mode is active by default unless

--economy

is set.

使用

references/command-interface.md

中的命令语法。

主要命令：

bash

no-slop --scan [target]
no-slop --fix [target]
no-slop --redesign [target]
no-slop --judge [target]
no-slop --prevent [target-or-brief]
no-slop --autopsy [target-or-brief]
no-slop -e --scan [target]

短标识：

```
-s
```
,
```
--scan
```
：扫描并评分。
```
-f
```
,
```
--fix
```
：精准整改。
```
-r
```
,
```
--redesign
```
：全面重新设计。
```
-j
```
,
```
--judge
```
：启动严格评审小组。
```
--prevent
```
：UI生成前的预防模式。
```
--autopsy
```
：生成AI UI法医式分析报告。
```
-e
```
,
```
--economy
```
：禁用评审子代理，仅使用本地确定性检查。

默认模式为

--scan

。除非设置

--economy

，否则评审模式默认启用。

First Action

第一步操作

Load

steps/step-00-command-router.md

Step 00 parses command flags, resolves dials and presets, loads the right references, and routes to scan, fix, redesign, judge, or prevent.

加载

steps/step-00-command-router.md

。

Step 00会解析命令标识，解析控制项和预设，加载对应参考资料，并路由至扫描、修复、重新设计、评审或预防流程。

Core References

核心参考资料

Load only what the task needs:

```
references/command-interface.md
```
: CLI grammar, flag precedence, output contracts.
```
references/dials-and-presets.md
```
: the eight 0-10 control dials and fast presets.
```
references/ai-slop-patterns.md
```
: detection taxonomy, severity scoring, slop signatures.
```
references/judge-system.md
```
: mandatory multi-agent judge protocol and fail gates.
```
references/prevention-protocol.md
```
: pre-generation detector and judge workflow.
```
references/remediation-playbook.md
```
: fix/redesign tactics and before/after examples.
```
references/design-quality-rubric.md
```
: scoring rubric and quality gates.
```
references/ecosystem-contract.md
```
: ownership, composition, and handoff rules.

仅加载任务所需内容：

```
references/command-interface.md
```
：CLI语法、标识优先级、输出约定。
```
references/dials-and-presets.md
```
：8个0-10级控制项和快速预设。
```
references/ai-slop-patterns.md
```
：检测分类、严重程度评分、敷衍特征。
```
references/judge-system.md
```
：强制多代理评审协议和准入门槛。
```
references/prevention-protocol.md
```
：生成前检测器和评审工作流。
```
references/remediation-playbook.md
```
：修复/重新设计策略及前后示例。
```
references/design-quality-rubric.md
```
：评分准则和质量门槛。
```
references/ecosystem-contract.md
```
：所有权、组合及交接规则。

Non-Negotiables

不可协商规则

Run 4-6 independent judge subagents by default for every non-economy command.
If subagent tools are unavailable in non-economy mode, mark
```
JUDGE_PANEL_UNAVAILABLE
```
; never claim a full pass.
Do not soften judge feedback with praise padding.
Do not replace one generic aesthetic with another.
Do not ship purple-blue gradients, glass cards, card soup, vague copy, fake proof, motion spam, token chaos, or brandless layouts unless the product context specifically justifies them.
Do not generate UI in
```
--prevent
```
until the detector and judge panel have approved a design contract.
Do not finish
```
--fix
```
or
```
--redesign
```
with failed judge gates. Iterate. If the gate still fails, block the result and state why.
At the end of
```
--fix
```
or
```
--redesign
```
, load the sibling skill
```
skills/reality-skill/SKILL.md
```
when the target includes an app, dashboard, form, CRUD flow, onboarding, checkout, settings, auth-like flow, fake proof, fake data, or primary actions. Use it to remove demoware without broad redesign.
Preserve behavior, accessibility, semantic structure, focus states, keyboard access, and responsive stability.
Use existing design tokens and component conventions when they are coherent. Replace them only when they are the problem.
For operational tools, prefer dense, scan-friendly, restrained interfaces. For expressive products, create a deliberate art direction, not decoration.

默认情况下，每个非经济型命令需运行4-6个独立评审子代理。
若非经济型模式下子代理工具不可用，标记为
```
JUDGE_PANEL_UNAVAILABLE
```
；绝不能声称完全通过。
不要用赞美之词弱化评审反馈。
不要用一种千篇一律的美学风格替代另一种。
除非产品场景明确需要，否则不得使用紫蓝渐变、毛玻璃卡片、卡片堆砌、模糊文案、虚假证明、过度动效、混乱令牌或无品牌布局。
在
```
--prevent
```
模式下，除非检测器和评审小组批准设计约定，否则不得生成UI。
若未通过评审门槛，不得完成
```
--fix
```
或
```
--redesign
```
流程。需反复迭代；若仍无法通过，需拦截结果并说明原因。
在
```
--fix
```
或
```
--redesign
```
流程结束时，若目标包含应用、仪表盘、表单、CRUD流程、引导页、结账流程、设置、类认证流程、虚假证明、虚假数据或核心操作，需加载关联技能
```
skills/reality-skill/SKILL.md
```
，用于移除演示性内容，无需进行大范围重新设计。
保留交互行为、无障碍特性、语义结构、焦点状态、键盘操作和响应式稳定性。
若现有设计令牌和组件约定连贯，需予以保留。仅当它们本身存在问题时才进行替换。
对于运维工具，优先选择紧凑、易扫描、克制的界面；对于创意产品，需打造明确的艺术风格，而非单纯装饰。

Required State

必需状态

Maintain this state throughout the workflow:

text

command:
target:
mode:
economy_mode:
preset:
dials:
slop_score_before:
slop_score_after:
slop_signatures:
autopsy_report:
issues_by_severity:
design_direction:
judge_panel:
judge_gate:
files_changed:
verification:
reality_handoff:
reality_changes:

在整个工作流中需维持以下状态：

text

command:
target:
mode:
economy_mode:
preset:
dials:
slop_score_before:
slop_score_after:
slop_signatures:
autopsy_report:
issues_by_severity:
design_direction:
judge_panel:
judge_gate:
files_changed:
verification:
reality_handoff:
reality_changes:

Local CLI Helper

本地CLI助手

For a deterministic local scan, run:

bash

python no-slop/scripts/no_slop_cli.py --scan <target>
python no-slop/scripts/no_slop_cli.py --autopsy <target>
python no-slop/scripts/no_slop_cli.py --prevent "brief text" --autopsy
python no-slop/scripts/no_slop_cli.py --preset dashboard --dial STRICTNESS=10 --scan <target>

The CLI supports context-calibrated scoring, dials, presets, signatures, economy mode, autopsy reporting, and a deterministic judge-gate summary. It treats regex matches as candidate evidence, then adjusts by confidence, counter-evidence, and nearby UI context. The Codex skill workflow remains the authority for actual code edits and live subagent judging.

如需进行确定性本地扫描，运行：

bash

python no-slop/scripts/no_slop_cli.py --scan <target>
python no-slop/scripts/no_slop_cli.py --autopsy <target>
python no-slop/scripts/no_slop_cli.py --prevent "brief text" --autopsy
python no-slop/scripts/no_slop_cli.py --preset dashboard --dial STRICTNESS=10 --scan <target>

该CLI支持上下文校准评分、控制项、预设、特征识别、经济型模式、分析报告以及确定性评审门槛汇总。它将正则匹配作为候选证据，再根据置信度、反证及周边UI上下文进行调整。Codex技能工作流仍是实际代码编辑和实时子代理评审的权威标准。