Receber por email
← Voltar para o catálogo

progressive-disclosure

Refactor large DataRobot skill files by moving detailed content into directly linked reference files while preserving meaning. Use when a skill triggers context-window warnings, needs progressive disclosure, or should be chunked without changing guidance.

20estrelas
Atualizado há 4 meses

Ver no GitHub ↗ Licença: Apache-2.0

Como adicionar

/plugin marketplace add datarobot-oss/datarobot-agent-skills

O comando exato pode variar conforme o repositório. Confira o README no GitHub.

Para o autor da skill

Cole no README do seu repo

Mostra que sua skill está catalogada na Skillteca, gera backlink e tráfego rastreável.

[![Skillteca](https://www.skillteca.com.br/api/badge/progressive-disclosure/svg)](https://www.skillteca.com.br/skills/progressive-disclosure)
#ai

Skills relacionadas

Ver todas de Desenvolvimento →

Alerta por categoria

Receba novas skills de Desenvolvimento toda segunda

1 email curto, só com as skills novas de Desenvolvimento. 4 minutos de leitura, sem spam, cancela com 1 clique.

Você confirma o email no primeiro envio. Sem spam. Cancela com 1 clique.

Ler descrição completa

Progressive Disclosure

Use this repo-local meta skill to split large skills into a concise SKILL.md plus adjacent reference files. This guides content moves and reference creation; it is not a content rewrite pass.

Guardrails

  • Do not rewrite, reinterpret, simplify, or change the meaning of skill content.
  • Do not delete details because they seem verbose. Move details to reference files instead.
  • Do not change SDK guidance, commands, workflows, safety notes, or examples except to preserve links after moving content.
  • Do not edit plugin manifests, versions, CODEOWNERS, or unrelated repo metadata unless the user explicitly asks.
  • Do not create deeply nested references. Reference files should be directly linked from SKILL.md.
  • Do not move first-use trigger guidance, required safety constraints, or critical prerequisites out of SKILL.md.
  • If content needs substantive editing, stop and ask the user before making that change.

Target Shape

Keep SKILL.md focused on:

  • Frontmatter with accurate name and description
  • When to use the skill
  • Quick start or primary workflow
  • Key decision points and safety constraints
  • Direct links to reference files

Move long supporting content into files such as:

  • reference.md
  • examples.md
  • troubleshooting.md
  • platform-setup.md
  • sdk-operations.md
  • ci-cd-patterns.md

Choose names that match the moved section. Prefer a small number of meaningful files over many tiny fragments.

Use folders only when the supporting material has a clear type or grouping:

  • references/ for longer conceptual guides, examples, and workflow details
  • scripts/ for executable helpers that the skill may ask the agent to run
  • Language folders such as python/, typescript/, or r/ when a skill supports multiple programming languages

Keep these folders directly under the skill directory. SKILL.md should link directly to files inside them, such as references/workflow-details.md, without requiring the agent to follow a chain of references.

Workflow

  1. Inspect the target skill:
  2. Read skills/<skill-name>/SKILL.md.
  3. Check current line count and major headings.
  4. Check current token warning/error status from task test:integration when available.
  5. Identify sections that are detailed reference material, long examples, command recipes, troubleshooting tables, platform-specific variants, or API details.
  1. Propose the split before editing when the change is substantial:
  2. List which sections will stay in SKILL.md.
  3. List which sections will move and the destination reference file for each.
  4. Keep a simple traceability map: original heading -> destination file -> replacement link text.
  5. Confirm any ambiguous sections with the user.
  1. Move content with minimal transformation:
  2. Preserve headings and body text as much as possible.
  3. Adjust heading levels only when needed for valid Markdown structure.
  4. Preserve code blocks exactly unless a relative link/path must change.
  5. Keep warnings, prerequisites, and safety notes attached to the relevant content.
  1. Replace moved content with concise pointers:

``markdown For detailed deployment configuration, see [deployment-reference.md](deployment-reference.md). ``

Each pointer should explain when to read the reference, not summarize the whole reference.

  1. Keep links one level deep:
  2. SKILL.md should link directly to every reference file the agent may need.
  3. Reference files may link back to SKILL.md if useful.
  4. Avoid chains like SKILL.md -> references/overview.md -> references/deep-reference.md.
  1. Validate:
  2. Run task test:integration for structural checks.
  3. Run task lint when tooling is available.
  4. If the goal is reducing context-window warnings, rerun the relevant tests and compare warning output. This repo warns at an estimated 3300 tokens and errors at 6700 tokens.

Candidate Section Heuristics

Good move candidates:

  • Long examples that are not needed for initial task routing
  • Repeated setup variants by platform or provider
  • Detailed SDK method inventories
  • Troubleshooting catalogs
  • Extended CI/CD snippets
  • Long validation or debugging procedures

Usually keep in SKILL.md:

  • The first-pass decision tree
  • Required safety constraints
  • Minimal happy-path workflow
  • Short command checklist needed for most runs
  • Links to all supporting references

Diff Review Checklist

Before finishing, verify:

  • Every moved block still exists in a reference file.
  • SKILL.md links directly to each new reference file.
  • The frontmatter description still includes clear trigger scenarios.
  • The final response includes the traceability map for reviewer confidence.
  • No instruction, recommendation, or example changed.
  • No unrelated formatting churn was introduced.
  • Repo validation passes or any local tooling gap is clearly reported.
Compartilhar X LinkedIn