Menü
Refactor existing AGENTS.md to follow progressive disclosure principles
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
| name | refactor-agents |
| description | Refactor existing AGENTS.md to follow progressive disclosure principles |
| allowed-tools | Read Write Edit Grep Glob AskUserQuestion Skill |
Optimize an existing AGENTS.md file using progressive disclosure principles. Extracts essentials to root, groups related instructions into sub-files, and flags content for deletion.
AGENTS_FILES = Glob("**/AGENTS.md")
ROOT_AGENTS = Read("AGENTS.md")
If no AGENTS.md exists: suggest Skill("init") instead.
Analyze ROOT_AGENTS for conflicting instructions:
| Contradiction Type | Example |
|---|---|
| Version conflicts | "Use Node 18" vs "Requires Node 20" |
| Style conflicts | "Use tabs" vs "2-space indent" |
| Tool conflicts | "Use npm" vs "pnpm install" |
| Process conflicts | "Always write tests first" vs "Tests optional for prototypes" |
For each contradiction found:
AskUserQuestion:
question: "Found conflicting instructions: [A] vs [B]. Which should we keep?"
options: [A, B, "Keep both with context", "Remove both"]
Extract ONLY content that belongs in root AGENTS.md:
| Keep in Root | Move to Sub-file |
|---|---|
| One-sentence project description | Detailed architecture docs |
| Package manager (if not npm) | Framework-specific patterns |
| Non-standard build/test commands | Testing conventions |
| Universal security rules | Component-specific rules |
| Links to sub-files | Verbose examples |
Root target: 60-100 lines (max 200)
Categorize instructions into logical groups:
| Category | Typical Content |
|---|---|
conventions/typescript.md | TS strict mode, type patterns, import style |
conventions/testing.md | Test structure, mocking rules, coverage |
conventions/git.md | Branch naming, commit format, PR process |
conventions/api.md | Endpoint patterns, error handling, auth |
conventions/styling.md | CSS approach, component styling, themes |
For ambiguous content:
AskUserQuestion:
question: "Where should '[instruction]' live?"
options: [category suggestions based on content]
Identify instructions that should be removed:
| Flag | Reason | Example |
|---|---|---|
| REDUNDANT | Agent already knows | "Use meaningful variable names" |
| VAGUE | Not actionable | "Write clean code" |
| OBVIOUS | No new information | "Test your changes" |
| OUTDATED | No longer applies | References deprecated tools |
| DUPLICATE | Already stated elsewhere | Same rule in multiple places |
Present flagged items for confirmation:
AskUserQuestion:
question: "These instructions seem [redundant/vague/obvious]. Delete them?"
options: ["Delete all", "Review each", "Keep all"]
Create the refactored file structure:
AGENTS.md # Minimal root (60-100 lines)
.agents/ # Sub-files directory
conventions/
typescript.md
testing.md
git.md
patterns/
[domain-specific].md
# [Project Name]
[One sentence description]
## Quick Reference
- **Stack:** [tech stack summary]
- **Build:** `[build command]`
- **Test:** `[test command]`
- **Lint:** `[lint command]`
## Conventions
See detailed guides in `.agents/conventions/`:
- [TypeScript](.agents/conventions/typescript.md)
- [Testing](.agents/conventions/testing.md)
- [Git Workflow](.agents/conventions/git.md)
## Security
- Never commit secrets or `.env` files
- [Other universal security rules]
## Before PR
\`\`\`bash
[single command that runs all checks]
\`\`\`
mv AGENTS.md AGENTS.md.backup.agents/ directory structureReport refactoring results:
## Refactoring Complete
### Before
- Root AGENTS.md: [X] lines
- Sub-files: [Y]
### After
- Root AGENTS.md: [X] lines (target: 60-100)
- Sub-files: [list with line counts]
### Changes
- Contradictions resolved: [count]
- Instructions moved: [count]
- Instructions deleted: [count]
- Redundant: [count]
- Vague: [count]
- Obvious: [count]
### Backup
Original saved to: AGENTS.md.backup
| Condition | Action |
|---|---|
| No AGENTS.md found | Suggest Skill("init") |
| Already minimal (<100 lines) | Ask if optimization still wanted |
| No contradictions found | Skip to Step 3 |
| User cancels at any step | Preserve original, no changes |
Create distinctive, production-grade frontend interfaces with high design quality. Use when building web components, pages, or applications. Includes framework-specific guidance for Tailwind, React, Vue, and Rails/Hotwire ecosystems.
Skill file structure, naming conventions, directory layout, frontmatter requirements, and invocation control. Use when creating skill files or slash commands to ensure correct format and validation.
Forces adversarial reasoning before committing to decisions. Triggers on architectural choices, approach selection, and planning phases to prevent premature commitment bias.
Verify LLM reasoning using Reverse Chain-of-Thought (RCoT) to detect overlooked conditions.
Autonomous task implementation - research, plan, build, review, fix, ship
Generate a complete favicon set (ICO, PNG variants, apple-touch-icon, web manifest) from a source image and integrate into the project's HTML layout. Use when user asks to generate favicons, set up PWA icons, or add an apple-touch-icon.