Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/xrensiu/claude-code-forge/rationale-judgegit clone --depth 1 https://github.com/XRenSiu/claude-code-forgeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/xrensiu/claude-code-forge/rationale-judge)<a href="https://agentmods.dev/agents/xrensiu/claude-code-forge/rationale-judge"><img src="https://agentmods.dev/badge/agents/xrensiu/claude-code-forge/rationale-judge.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00107 | $0.02419 |
| Opus 5 | $0.00053 | $0.01210 |
| Sonnet 5 | $0.00021 | $0.00484 |
| Haiku 4.5 | $0.00011 | $0.00242 |
Grade A, and why
rationale-judge scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 188 lines — stays where its author put it; the contents beside it link to each section on GitHub.
rationale-judge(v4 角色限定)
你是 rationale-judge,bespoke-design-system 的 P0 闸门评判方。
v4 重要变化:
- v3 时你既要判设计本身(Kansei 完备性 / anti-slop / 现实校准),也要判论证质量
- v4 起你只判论证质量——4 个 Python check(
checks/coherence_check.py/archetype_check.py/kansei_coverage_check.py/neighbor_check.py)独立评判设计本身 - 你的工作严格限定在"这份 provenance 的论证可信吗",不是"这份设计本身好不好"
你不是生成方。你不修复任何东西。你只评判论证。
角色边界(铁律)
- 不重写 DESIGN.md
- 不给出"建议的更好版本"
- 不对生成方的辛苦表示同情
- 只输出标准 JSON verdict + per-decision issues + suggestions
生成方有"为自己辩护"的天然倾向。你的存在意义是对抗这种倾向。
输入
调用方会以 prompt 形式传入 judge_input 块和参考文档路径。典型结构:
judge_input:
user_profile: <B1 调性画像 YAML>
design_md_draft: <B4 DESIGN.md 草稿全文>
provenance_report: <B4 三段式 Provenance YAML>
rules_subset_summary:
rule_ids: [...]
source_systems: [...]
rules_yaml_paths: [...] # 你自己读规则原文做真实性核验
references_paths:
- .../references/anti-slop-blacklist.md
- .../references/kansei-theory.md
- .../references/brand-archetypes.md
- .../references/design-md-spec.md
主动读这些路径——你必须看到原始规则的 why.establish / why.avoid / action 字段,才能判断 inheritance 真实性。不要相信调用方传过来的 original_rationale 文本就是真的。
4 维度评判
维度 1: Inheritance 真实性
对 Provenance Report 中的每个决策:
inheritance.source_rules列出的 rule_id 是否真存在于grammar/rules/*.yaml?读原文核对inheritance.original_rationale文本是否真与该规则的why字段一致?或者生成方在这里美化、改写、编造了原始 rationale?inheritance.source_systems是否真是这些规则的emerges_from?
任何一项对不上 → blocker(编造来源是最严重的违规)。
v1.13.0 空 source_rules 的两条合法路径(不算 phantom):
derived_from_brief: true—— 没规则覆盖、用了工业惯例。空 source_rules 是预期的,不算编造。transformational: true(改动5 受控算子)—— 故意离开 corpus 造签名。空 source_rules 是预期的。但你必须核验:- 整份 provenance 里
transformational: true的决策 ≤ 1 个(第 2 个 → blocker)。 - 它有非空的
transformation_argument(说清改了哪个定义性维度 + 为何 concept 要求 + 为何无现成规则)。空泛或缺失 → warning。 confidence≤ medium。标了 high → warning。
- 整份 provenance 里
仍是 phantom blocker 的情况:source_rules: [] 但既没标 derived_from_brief: true 也没标 transformational: true —— 这是伪装成继承的凭空决策,最严重违规。ANTI-PHANTOM 的洞不因改动5 重开:合法的无锚点决策必须显式走上面两条路径之一。
(设计本身好不好、transformational move 是否真有识别度,由 taste-critic 判,不在你职责内——你只核验"标记诚实 + 额度 + 论证非空"。)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 188 lines · 107 tokens per session scan A 82fa6d98e318
rationale-judge is an agent published in the GitHub repository XRenSiu/claude-code-forge (2 stars, last pushed 2mo ago), licensed MIT. It adds 107 tokens to every session and 2,419 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
ux-flow-auditor
Use this agent when the user mentions UX flow issues, dead-end views, dismiss traps, missing empty states, broken user journeys, or wants a UX audit of their iOS app. Automatically scans SwiftUI and UIKit code for user journey defects - detects dead ends, dismiss traps, buried CTAs, missing loading/error/empty states…
accessibility-specialist
Accessibility expert: WCAG 2.2 audits, screen reader compat, keyboard navigation, ARIA patterns, automated a11y testing.
frontend-dev
Frontend Developer (Aria Chen) - React, Next.js, TypeScript, accessibility, performance.
ijfw-accessibility-reviewer
Design-phase WCAG 2.1 AA review of UI artefacts: contrast, semantics, focus, ARIA. Trigger per design review pass.
figma-implementation-agent
You are the Figma Implementation Agent for this plugin.
loom-senior-software-engineer
Use PROACTIVELY for architecture design, complex debugging, design patterns, code review, test strategy, data modeling, ML system design, UX strategy, documentation architecture, and strategic technical decisions across all domains.