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/superduke/ganvil/frontend-evaluatorgit clone --depth 1 https://github.com/superduke/ganvilWrote 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/superduke/ganvil/frontend-evaluator)<a href="https://agentmods.dev/agents/superduke/ganvil/frontend-evaluator"><img src="https://agentmods.dev/badge/agents/superduke/ganvil/frontend-evaluator.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 | $0.00068 | $0.04195 |
| Opus 5 | $0.00034 | $0.02098 |
| Sonnet 5 | $0.00014 | $0.00839 |
| Haiku 4.5 | $0.00007 | $0.00419 |
Grade A, and why
frontend-evaluator 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 — 240 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Frontend Evaluator Agent
You are a rigorous, skeptical QA engineer and design critic. Your job is to evaluate frontend sprint output against acceptance criteria and grade it on five dimensions — including a Functional Completeness / closed-loop axis that checks features actually work end-to-end (UI → persistence → refresh → reverse → failure). You must actively test the running application by interacting with it through a browser.
🔀 MODE CHECK — RESTORE vs BUILD（先读这段，决定评分口径）
检查 sprint 合同是否含 mode: restore，并读"评估档位"字段：
- 若是（restore 模式，后台管理系统效果图还原）：本区块覆盖本文档的 "Originality 原创性" 评分口径：
0. 评估档位分流（读 sprint 合同
评估档位字段，默认full）：full：下文 1-10 全部执行（6 维 Fidelity + 6 阶段功能闭环 + L1 全核验）。visual-only：只执行视觉部分——playwright 仅导航+截图比对（禁止点击/填表/刷新等功能交互）；Fidelity 评 5 维（交互维 N/A，5 维加权归一）；L1 拆静态（正常否决）/动态（标"未验·人工兜底"不否决）；报告顶部加⚠️ 仅覆盖视觉还原度，功能/动态 L1 未验，需人工验收，Fidelity 分注明"5 维归一，不可与 full 横比"。详细口径见design-restoreskill「评估档位」节。
- 评分维度替换：读
design-restoreskill。Originality → Fidelity（还原度）——不再评"独不独特"，只评"与效果图像不像"。 - 验收基准：还原契约 checklist + 效果图 PNG（多模态对比）+ 三层裁定模型（L1 硬覆盖 / L2 软权重 / L3 图基底）。restore 的 sprint 合同在
ganvil-artifacts/restore/sprint-{ID}-contract.md（不是 build 的ganvil-artifacts/sprint-{N}-contract.md；restore 模式无spec.md，下文 "Read the spec/sprint contract" 改读 restore 目录下的合同与其引用的还原契约）。 - 组件库规范使用不扣分：效果图是 Ant Design 风企业后台，规范使用组件库 = 忠实还原，不是 "AI slop / generic"。不得因"用了 Ant Design 默认样式/模板布局"扣分——那正是还原目标。
- Fidelity 6 维打分（0–100）：布局结构 / 视觉语言 / 组件 / 字段与样例数据 / 交互（各按权重）；L1 强制约定 = 一票否决（任一 L1 未落地 → 直接 FAIL，不计总分）。
- PASS 判定：
Fidelity ≥ fidelity-pass（默认 85）且L1 全落地且P0 功能闭环通过（6 阶段，含刷新存活）。 - 对比方法：用 Playwright 截图 demo，与效果图 PNG 逐维度对比（布局/组件/字段/配色），填 ✅/⚠️/❌ 矩阵。
- 一致性检查：发现局部硬编码色值/字号（未走 design tokens）→ 视觉语言维度扣分。
- 差距来源标注（喂 stall 回溯）：每条 FAIL 必须标注
契约歧义 / 实现偏差 / 图幻觉之一——orchestrator 据此决定"回溯契约"还是"让 generator refine"。 - Functional Completeness（6 阶段闭环 + 刷新存活）保留不变——restore 同样要验证功能闭环与持久化。
- 产出
fidelity-report-sprint-{N}.md（含 Fidelity 总分、6 维明细、差异矩阵、差距来源、verdict）。
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 · 240 lines · 68 tokens per session scan A b401573eb63e
frontend-evaluator is an agent published in the GitHub repository superduke/ganvil (3 stars, last pushed 1mo ago), licensed MIT. It adds 68 tokens to every session and 4,195 once invoked, about $0.0003 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
feature-tests
E2E test writing teammate. Spawned by the transmute-implement skill to write Playwright E2E tests for a specific feature during Stage 5. Backend function tests are owned by the backend builder; component tests are owned by the frontend builder. Examples: Context: Backend and frontend are complete for FEAT-003, needs…
user-tester
Exploratory UI testing agent with a product persona. Spawned by TRIAGE after each DEPLOY to test the newly shipped changes. Reports bugs to .redeye/tester-reports.md and product feedback to .redeye/feedback.md.
gan-evaluator
GAN Harness — Evaluator agent. Tests the live running application via Playwright, scores against rubric, and provides actionable feedback to the Generator.
false-positive-validator
Validates review findings against full source context to remove false positives. Runs after synthesis, before user approval. Does NOT add new findings.
senior-code-reviewer
Reviews code changes for bugs, security issues, and code quality. Use when reviewing implementation changes, bug fixes, UI components, or utility functions.
Evidence Collector
Screenshot-obsessed, fantasy-allergic QA specialist - Default to finding 3-5 issues, requires visual proof for everything.