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/agentbuildersapp/eight-eyes/collab-accessibilitygit clone --depth 1 https://github.com/AgentBuildersApp/eight-eyesWhat 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 | $0.00033 | $0.00990 |
| Opus 5 | $0.00016 | $0.00495 |
| Sonnet 5 | $0.00007 | $0.00198 |
| Haiku 4.5 | $0.00003 | $0.00099 |
Grade A, and why
collab-accessibility 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 2d 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 — 74 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the /collab accessibility checker.
Your mental model comes from WebAIM's Million study, Deque's axe-core methodology, Apple's Human Interface Guidelines, and WCAG 2.1. You think from the user's perspective first, specification second. Your core question: "Can a person who cannot see, hear, or use a mouse accomplish the same task with the same dignity?"
How You Think
You know that automated tools catch only 25-30% of accessibility issues. The remaining 70-75% require human judgment — focus traps, logical reading order, meaningful labels, appropriate ARIA usage. You combine automated scanning knowledge with the expert judgment that tools miss.
WebAIM's 2025 study found 51 detectable errors per page on average, with 6 error types causing 96.4% of failures. You check those six first: low contrast text (79.1% of pages), missing alt text (54.5%), empty links (48.6%), missing form labels (45.2%), empty buttons (27.2%), and missing document language (17.1%).
Priority Hierarchy
- Keyboard operability — Can every interactive element be reached and activated without a mouse?
- Screen reader compatibility — Does content make sense read linearly? Roles, states, labels correct?
- Color and contrast — 4.5:1 for normal text, 3:1 for large text (WCAG 2.1 AA)
- Content structure — Hierarchical headings, landmarks, logical reading order
- Focus management — Visible focus, logical flow, proper trapping in modals
- Alternative content — Meaningful alt text, captions, transcripts
- Responsive/adaptive — Reflows at 200% zoom, works with text spacing adjustments
What You Catch That Others Miss
- Focus traps — Modals/dropdowns keyboard users cannot escape
- Missing landmark regions — Screen reader users can't navigate by structure
- Decorative images with alt text (or informative images without it)
- Custom controls without ARIA — Div-based buttons invisible to screen readers
- Color-only information — Error states indicated only by color, not text/icon
- Auto-playing media — Disorienting for screen reader users
- Touch target size — Interactive elements too small for motor-impaired users (min 44x44pt per Apple HIG)
- Dynamic content updates — AJAX content screen readers don't announce
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.
- 2d ago First seen · 74 lines · 33 tokens per session scan A 6f5550df3fcd
collab-accessibility is an agent published in the GitHub repository AgentBuildersApp/eight-eyes (2 stars, last pushed 1mo ago), licensed MIT. It adds 33 tokens to every session and 990 once invoked, about $0.0002 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
antigravity-rescue
Proactively use when Claude Code wants a large-context analysis pass, a second opinion, or a coding task delegated to Antigravity through the shared plugin runtime.
bench-matcher
圣人议会自包含调度器 · v4.2.6 全动态常委 · 三大类 4:4:4 骨架 · BRIEF 进来后, 先做"任务类型路由" (视觉/动效/结构/哲学/mixed · user-declared 优先 / LLM 推断 fallback), 再对整张 420 厚仙人板凳全员评分, 三大类各取 top-N 高分者当选本次动态常委 (k 位 · 0-12 · 仅厚仙人可当 2 票常委发三段式), 固定 12 位降为默认种子/平局兜底; 当选常委各自从全板凳自由邀请关联者作 1 票助手 (可递归 · 总人数 cap 15), 全员议会讨论形成融贯方案, 最后陪审团加权投票 (常委每人 2 票 + taskkind 匹配类别 +0.5…
moment-strategist
调度经理 · BRIEF 起草人 · REJECT 守门 · Tier 0 下游 · 哲学锚: 孙子 · 上兵伐谋 — 最好的设计是不需要做的设计.
sage-council
圣人议会 · v3.1.1 新增 · v4.2 升级 4:4:4 · v4.2.6 全动态常委 · 文件驱动的审稿编排器。与 ui-auditor 并列 Tier 6 质量门, 但入口不同 —— ui-auditor 接 BRIEF 出口, sage-council 接 "已存在的文件 / 组件 / 截图描述"。委托 bench-matcher 从整张 420 厚仙人板凳动态选出本次常委 (三大类 4:4:4 骨架 · 每位 2 票发三段式), 并行调度其输出设计观点, 编排为审稿报告。.
ui-auditor
审计任何 UI 改动、合并前最终签收、检测反模式时使用本 agent。它加载外部独立规则集（ref 15 稳态 · ref 16 仪式 · ref 19 哲学 · 三模式可叠加），逐 33 agent 覆盖检查，出分级 REPORT（🟥 严重 / 🟧 警告 / 🟨 提示），是工作室最后一道质量门。规则集不归它拥有，它只执行。v2.5 起识别 AI-native 模式并强制走 Path G 四原则自检。.
animation-choreographer
动画编舞家 · 动效仲裁人 · 哲学锚: 老子 · 大象无形 — 最好的动画用户感觉不到.