メニュー
Create, validate, audit, review, refactor, optimize, and improve Agent Skills. Works with YAML frontmatter, SKILL.md docs, trigger-rich descriptions, slash commands, and automated validation + grading to keep skills consistent and discoverable.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | skill-helper |
| description | Create, validate, audit, review, refactor, optimize, and improve Agent Skills. Works with YAML frontmatter, SKILL.md docs, trigger-rich descriptions, slash commands, and automated validation + grading to keep skills consistent and discoverable. |
| license | MIT |
| metadata | {"version":"2.0.0","author":"OpenCode","replaces":["skill-creator","skill-validator"],"audience":"developers, agents","workflow":"skill-development"} |
Prerequisites:
Tools Used: Read, Write, Edit, Bash, Grep, Glob
Three Primary Workflows:
Create a New Skill
User: "Create a skill for [task]"
Agent: Follows 8-step creation workflow
Validate & Grade Existing Skill
./scripts/validate-skill.sh path/to/skill
# Returns: 34 validation checks + 100-point grade
See: references/validation-guide.md
Rubric: references/grading-rubric.md
Optimize for Discoverability
User: "Make [skill-name] more discoverable"
Agent: Analyzes description, adds verbs/keywords
Use this skill when you:
Auto-Activation: 5+ action verbs, 10+ keywords, "Use when" triggers, parenthetical grouping, task-centric framing
Progressive Loading: L1 (frontmatter ~100 tokens) → L2 (SKILL.md 1000-2000 tokens) → L3 (references/ unlimited)
Token Targets: Sweet spot 1000-2000 tokens; warn <500 or >3500
Validation + Grading: 34 checks + 100-point rubric; grade calculated first, used as validation input
Run: ./scripts/validate-skill.sh path/to/skill
33 checks across 3 tiers:
See references/validation-guide.md for complete details.
100-point rubric: Description Quality (35%), Structure (25%), Content (20%), Technical (15%), Compliance (5%)
Grades: A (90-100), B (80-89), C (70-79), D (60-69), F (0-59)
See references/grading-rubric.md for scoring details.
Workflow Skills: Multi-step processes with phases (planning, deployment) Reference Skills: Domain knowledge with quick references (testing, security) Tool Skills: Specific operations with clear input/output (validation, formatting)
See references/creation-workflow.md for pattern details.
| Element | Rule | Example |
|---|---|---|
| Name | ^[a-z0-9-]+$ (max 64) | spring-boot-core |
| Description | 50-1024 chars, 5+ verbs, 10+ keywords | See frontmatter above |
| Tokens | 1000-2000 ideal, >500 min, <3500 warn | ~200-400 lines |
| Sections | What I Do, When to Use Me, Quick Start | Required |
| Resources | assets/, references/, scripts/ | Optional |
| Command | {name}.md or ak-{name}.md | Matches skill name |
| Grade | A-F scale (100 points) | Target: B+ (85+) minimum |
Create new skill:
User: "Create a skill for API validation"
→ 8-step workflow → Grade B+ (87/100)
Validate existing skill:
./scripts/validate-skill.sh .opencode/skill/my-skill
→ 0 errors, Grade A (92/100)
Optimize description:
User: "Make skill more discoverable"
→ Add verbs + keywords → Grade improves C (75) to A (93)
See references/creation-workflow.md for detailed examples.
| Error | Solution |
|---|---|
| "Invalid name format" | Use lowercase, hyphens only |
| "Description too short" | Add action verbs + keywords |
| "SKILL.md too large" | Move content to references/ |
| "No slash command" | Create command file |
See references/validation-guide.md for complete error catalog.
Load detailed guidance as needed:
| Reference | Use When | Content |
|---|---|---|
| creation-workflow.md | Creating new skills | 8-step process, patterns, examples |
| validation-guide.md | Understanding validation | 34 checks explained, fixes |
| optimization-patterns.md | Improving descriptions | Trigger-rich patterns, testing |
| grading-rubric.md | Understanding grades | 100-point scoring breakdown |
markdown-editor - For creating and editing SKILL.md markdown filesgithub-actions - For CI/CD integration of validationgithub-project-manager - For tracking skill development tasksComplete workflows & resources:
Templates:
Automation:
./scripts/validate-skill.sh - Validation + grading in one commandRun, write, generate, record, save, persist, and synthesize agent retrospectives and lessons learned after completing tasks. Self-assess session quality, identify mistakes, extract reusable patterns, and update persistent LESSONS.md files using the retrieve-verify loop. Auto-trigger setup: install OpenCode plugin hooks, Claude Code Stop hooks, and Cline TaskComplete hooks so retros fire automatically when all todos complete. Works with session_read (OpenCode), context-window reconstruction (Claude Code, Cursor, Gemini CLI), and compaction pipelines. Use when finishing a task, at session end, when asked to reflect, review, debrief, or write a post-mortem, or when setting up automatic retro triggers.
Create, build, edit, validate, run, and export AnyLogic models with local-binary workflows and documented model structures (statecharts, parameter expressions, experiments, CLI, XML, Java, ALP, ALPX). Use when building or revising AnyLogic statecharts, changing parameter expressions, probing a local installation, running experiments, exporting standalone applications, or checking which documented binary surfaces are actually available.
Create, build, refactor, and optimize Angular components with signals, inputs, outputs, lifecycle hooks, change detection, content projection, and smart/presentational patterns. Use when building UI components, fixing component bugs, or improving performance.
Create, implement, debug, refactor, and optimize Angular state management with RxJS observables, subjects, signals, operators, and reactive patterns. Fix memory leaks, manage subscriptions, handle async data streams. Use when building reactive Angular components, services, or stores.
Write, create, generate, debug, and fix Angular unit tests, component tests, service tests, e2e tests with Vitest, Jest, Jasmine, TestBed, Cypress, and Playwright. Use when testing components, services, signals, observables, DOM interactions, or mocking dependencies.
Create, configure, optimize, and debug GitHub Actions CI/CD workflows, jobs, triggers, caching, matrix builds, reusable workflows, and secrets. Use when writing .github/workflows YAML, fixing pipeline failures, or improving CI speed.