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

Score Loop

ASecurity

Run a demanding task through a generator-evaluator loop with a cutoff grade — the agent builds at high performance standard, an evaluator tests and grades the result against a predefined weighted rubric, and if the score is below the threshold the work is redone and re-graded until it passes or the iteration budget is exhausted. Use when a task must reach a minimum quality score (e.g. 85/100) before it is considered finished. Also covers warm-start: persist the winning strategy ("sabão venced...

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

Security Analysis

A100/100

Scanned 9/19/2026

Install to Claude Code

$npx -y skills add majinmagros/magros.ai-skills --skill score-loop --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Score Loop?

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

Security grade badge for Score Loop
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majinmagros-score-loop/badge)](https://www.skillsdirectory.com/skills/majinmagros-score-loop)

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

Download Zip
Files
SKILL.md
---
name: score-loop
description: Run a demanding task through a generator-evaluator loop with a cutoff grade — the agent builds at high performance standard, an evaluator tests and grades the result against a predefined weighted rubric, and if the score is below the threshold the work is redone and re-graded until it passes or the iteration budget is exhausted. Use when a task must reach a minimum quality score (e.g. 85/100) before it is considered finished. Also covers warm-start: persist the winning strategy ("sabão vencedor") so the next run of the same task-signature starts from the winner instead of retesting soaps that already failed. Triggers on "sabão vencedor", "reusar a estratégia que funcionou", "warm start", "não repetir os que falharam", "começar da última iteração vencedora", "loop com memória".
argument-hint: "<task> --nota-corte <0-100> --max-iteracoes <N>"
---

# Score Loop

Proof-by-grading: build, grade, iterate until a minimum score is met.

## When to use

When the result must be proven reliable before finalizing — demanding coding tasks, refactors, migrations, or any delivery where "good enough" is decided by a numeric grade against predefined criteria.

## Steps

0. **Warm-start (recuperar vencedor)** — antes de iterar, derive a assinatura da
   tarefa (ex.: tipo + escopo + critério principal) e procure um registro
   vencedor em `.ecc/vencedores.json`. Se existir, comece a rodada **na
   estratégia vencedora** (prompt/parâmetros registrados) com pequena
   perturbação em volta — não re-teste os sabões que falharam. Se não existir,
   siga do zero.
1. **Define the contract** — before writing anything, state:
   - Task (the deliverable)
   - Cutoff score (default **85** of 100; adjustable via argument)
   - Weighted rubric (default): Funcionalidade 0.4 · Confiabilidade 0.3 · Manutenibilidade 0.2 · Clareza 0.1
   - Iteration budget (default 5)
2. **Challenging prompt** — approach the task at maximum performance standard: demand the best possible solution, and explicitly list what would count as "not acceptable" (broken edge cases, unhandled errors, hidden assumptions, untested paths).
3. **Build** — implement the deliverable against the contract.
4. **Evaluate** — switch explicitly to the evaluator role (never grade in the same breath as building): test the result for real — run tests, execute code paths, probe edge cases — and score each rubric criterion 0-100. Final grade = weighted sum. If a separate agent or subagent is available, delegate evaluation to it for independence.
5. **Gate** — compare final grade to cutoff:
   - `grade >= cutoff` → finish with the report and **register the winner** (step 7)
   - `grade < cutoff` → write specific per-criterion feedback (what exactly missed points), then return to step 3 and rebuild addressing every point
6. **Plateau guard** — if the iteration budget is exhausted or the grade stagnates (delta < 2 points across 2 consecutive iterations), stop and report the blocker: NEVER auto-approve below the cutoff.

## Warm-start ("sabão vencedor")

A analogia: uma máquina de lavar que testa vários sabões até a mancha sair. No
4º loop o sabão X funcionou — semana que vem, use direto o X, sem re-testar os
que falharam.

### Registro vencedor (schema `.ecc/vencedores.json`)

```json
{
  "task-signature": "gerar-thumbnails-variantes",
  "strategy": {
    "prompt": "...",        // a estratégia reprodutível (prompt/parâmetros/abordagem)
    "params": { ... }       // ex.: model, temperature, nota-corte, top_k
  },
  "score": 92,
  "iteracoes": 4,
  "timestamp": "2026-08-17T12:00:00Z"
}
```

- Path padrão: `.ecc/vencedores.json` no workspace; configurável via env
  `WARM_START_FILE`.
- Assinatura da tarefa deve ser estável entre rodadas (mesmo tipo + escopo),
  senão o warm-start não encontra o vencedor.
- **Só persiste em PASS verificado** (grau >= corte com verificador
  independente). NUNCA persista best-effort.
- Rodada nova que supera o vencedor registrado → **substitui**.
- Revalidar periodicamente: estratégia vencedora pode degradar se o modelo ou o
  problema mudar. Se o warm-start parar de passar, reverta ao loop do zero.

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 →