Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsCommunityBlog
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

Back to skills

Knowledge Work Proxy Eval

ASecurity

Evaluate agents on hard knowledge-work task packs (banking, consulting, legal) as a proxy for your domain. Use when testing agents on vague prompts plus document workspaces. Use quando avaliar agentes com pacotes banking/consulting/legal, prompts vagos realistas e workspace de documentos.

2 stars
0 votes
0 copies
0 views
Added 9/19/2026
ai-agentsgotesting

Security Analysis

A100/100

Scanned 9/19/2026

Install to Claude Code

$npx -y skills add majinmagros/magros.ai-skills --skill knowledge-work-proxy-eval --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Knowledge Work Proxy Eval?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Knowledge Work Proxy Eval
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majinmagros-knowledge-work-proxy-eval/badge)](https://www.skillsdirectory.com/skills/majinmagros-knowledge-work-proxy-eval)

More formats (shields.io, HTML) on the badges page.

Download Zip
Files
SKILL.md
---
name: knowledge-work-proxy-eval
description: Evaluate agents on hard knowledge-work task packs (banking, consulting, legal) as a proxy for your domain. Use when testing agents on vague prompts plus document workspaces. Use quando avaliar agentes com pacotes banking/consulting/legal, prompts vagos realistas e workspace de documentos.
---

# Knowledge Work Proxy Eval

Se o agente resolve banking/consulting/legal duro, provavelmente resolve o seu domínio. Avalie por proxy, não por demo.

## Quando usar

Quando precisa saber se um modelo/harness aguenta trabalho de conhecimento real (documentos + ambiguidade + deliverable), antes de confiar no seu domínio. NÃO é execução de pesquisa — é avaliação.

## Passos

1. **Escolha o proxy duro** — banking (investment analysis), consulting (market sizing, segmentação), legal (regulatory, M&A, compliance). Duro de propósito: se passar aqui, transfere para domínios mais fáceis.
2. **Monte o pacote de task** — cada task tem 4 partes:
   - Prompt curto e vago como usuário real escreveria.
   - Workspace de documentos (planilhas, charts, PDFs, estrutura de pastas real).
   - Ferramentas permitidas e formato do deliverable esperado.
   - Trajectory esperada: quais arquivos ler, quais contas fazer, em que ordem.
3. **Exija trajetória, não só resposta** — o agente deve ler o workspace, cruzar fontes e mostrar o caminho (arquivos lidos, tool calls, cálculos). Resposta certa com trajetória errada = fail.
4. **Grade com régua de expert** — gabarito criado/validado por experts do domínio. Score por acurácia do deliverable + fidelidade da trajetória. Acompanhe saturação: ~85-90%+ = benchmark saturado, troque de pacote.
5. **Construa o proxy do SEU domínio** — clone a anatomia: 5-10 tasks reais, prompts vagos coletados de usuários, workspace anonimizado, gabarito com tolerância numérica explícita. Comece com 3 tasks piloto antes de escalar para 50.
6. **Decida a stack** — reporte por modelo: acurácia no proxy + custo + tempo. SOTA para tasks críticas, workhorse para volume, leve só com supervisão. Um benchmark isolado nunca decide.

## Regras

- NEVER avaliar knowledge work só com pergunta-resposta sem workspace: sem documentos não há proxy.
- NEVER aceitar número sem memória de cálculo e fonte.
- NEVER confundir com execução: `deep-research`/`market-research` EXECUTAM pesquisa; este skill AVALIA o executor.
- Prompts do pacote ficam vagos de propósito; não "ajude" o agente no enunciado.
- Tasks impossíveis ou ambíguas sem saída "não sei" medem honestidade, não acurácia — grade separado.

## Related skills

- `agent-eval` — harness de avaliação first-party.
- `deep-research` — execução de pesquisa profunda (objeto da avaliação, não o juiz).
- `agentic-benchmark-top5` — o índice pessoal onde este proxy mora.

Attribution

majinmagrosmajinmagros
View sourceMore from majinmagros →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

1023331 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

686011 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3331 votes

catchup

Recovers prior coding-agent session context by running `catchup <agent> --since-compact`, which extracts a clean summary of a previous Codex, Claude Code, Antigravity, OpenCode, or Pi Agent session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", or asks to recover/summarize a previous session before continuing. Do NOT use for the current conversation, git history, or any non-agent log.

611 votes

math-skill

A comprehensive mathematical reasoning skill for AI assistants — handles arithmetic to research-level problems with rigorous step-by-step reasoning, systematic verification, and transparent uncertainty handling

381 votes
View all in ai-agents →