AgentSkillsCN

opencode-improved

分析外部资源(GitHub仓库、文档、文章),并为OpenCode配置提出改进建议。无论是发现新规律、对比不同实现,还是强化现有技能、命令与代理,此功能都能助你事半功倍。 当用户提出“基于……进行改进”、“从……中学习”、“与……进行对比”、“我能从……中学到什么”,或提供链接供你分析时,应主动运用此功能。 示例: - 用户输入:“我能从这个仓库中学到什么?” → 分析并提出切实可行的改进建议 - 用户输入:“根据这篇文章改进我的代理” → 提取其中的模式,提出更新方案 - 用户输入:“/improve https://github.com/example/config” → 对比并推荐优化方案

SKILL.md
--- frontmatter
name: opencode-improved
description: |-
  Analyze external resources (GitHub repos, documentation, articles) and suggest improvements to OpenCode configuration. Use for discovering new patterns, comparing implementations, and enhancing existing skills, commands, and agents.
  
  Use proactively when user says "improve based on", "learn from", "compare to", "what can I learn from", or provides a link to analyze.
  
  Examples:
  - user: "What can I learn from this repo?" → analyze and suggest concrete improvements
  - user: "Improve my agents based on this article" → extract patterns, propose updates
  - user: "/improve https://github.com/example/config" → compare and recommend enhancements
<overview>

Analyze external resources and suggest actionable improvements to OpenCode configuration.

This skill enables systematic improvement by:

  1. Analyzing external resources (repos, docs, articles)
  2. Extracting relevant patterns and techniques
  3. Comparing against current configuration
  4. Suggesting concrete, actionable improvements
</overview> <workflow> <phase name="resource-acquisition">

Phase 1: Resource Acquisition

  1. Identify resource type:

    • GitHub repository → Fetch relevant config files
    • Documentation URL → Extract key concepts
    • Article/blog → Identify patterns and recommendations
  2. Fetch content:

    • For GitHub: Look for opencode.json, .opencode/, AGENTS.md
    • For docs: Extract configuration examples
    • For articles: Identify actionable recommendations
</phase> <phase name="pattern-extraction">

Phase 2: Pattern Extraction

MUST extract patterns in these categories:

CategoryLook For
AgentsNew archetypes, permission patterns, tool configs
SkillsOrganization patterns, reference structures
CommandsWorkflow shortcuts, argument patterns
ConfigProvider setups, MCP configurations, permissions
</phase> <phase name="gap-analysis">

Phase 3: Gap Analysis

MUST compare extracted patterns against:

  1. Existing .opencode/ structure
  2. Current opencode.json configuration
  3. Defined agents, skills, and commands
</phase> <phase name="recommendations">

Phase 4: Recommendations

MUST generate recommendations with:

  • Priority: High / Medium / Low
  • Effort: Minimal / Moderate / Significant
  • Impact: Description of benefit
  • Implementation: Concrete steps or code
</phase> </workflow> <output-format>

Output Format

markdown
## Analysis Summary

**Source**: [URL or repo name]
**Type**: [Repository | Documentation | Article]
**Relevance Score**: [High | Medium | Low]

## Key Findings

### Finding 1: [Title]

**Pattern**: [What they do]
**Current State**: [What you have]
**Gap**: [What's missing]

### Finding 2: [Title]
...

## Recommendations

### 1. [Recommendation Title] (Priority: High, Effort: Minimal)

**Impact**: [What this improves]

**Implementation**:
\`\`\`jsonc
// Code or configuration example
\`\`\`

### 2. [Recommendation Title] (Priority: Medium, Effort: Moderate)
...

## Not Applicable

These patterns from the source don't apply because:
- [Pattern]: [Reason it doesn't fit]
</output-format> <rules>

Evaluation Criteria

Worth Adopting

  • SHOULD solve a problem you currently have
  • SHOULD improve developer experience
  • SHOULD reduce repetitive configuration
  • MUST follow OpenCode best practices
  • MUST be compatible with existing setup

Skip If

  • Over-engineered for your use case
  • Conflicts with existing patterns
  • Requires major restructuring for minimal benefit
  • Outdated or deprecated approaches
  • Not applicable to your workflow
</rules> <guidelines>

Opinionated Guidance

This skill is intentionally opinionated:

  • Honesty over politeness: If nothing is useful, say so
  • Quality over quantity: Fewer good suggestions > many mediocre ones
  • Practicality over theory: Focus on actionable improvements
  • Compatibility matters: MUST NOT suggest breaking changes

Response Templates

When useful patterns found:

"I found 3 patterns worth adopting and 2 that don't fit your setup. Here's what I recommend..."

When nothing applicable:

"After analyzing [resource], I don't see improvements that would benefit your current setup. The patterns there are [too basic / too specialized / incompatible] because [reason]."

When partially useful:

"One pattern from [resource] could help: [description]. The rest is either already covered or not applicable."

</guidelines> <examples>

Integration with Slash Command

This skill works with the /improve command:

markdown
---
description: Analyze a resource and suggest OpenCode improvements
---

Analyze the following resource and suggest improvements to my OpenCode configuration:

$ARGUMENTS

Use the opencode-improved skill. Be opinionated and honest about what's useful.
</examples> <constraints>

Best Practices

MUST

  • Fetch and analyze the actual content, not guess
  • Compare against current configuration specifically
  • Provide implementation examples, not just suggestions
  • Be honest when nothing is useful
  • Prioritize by impact and effort

MUST NOT

  • Suggest changes without understanding current setup
  • Recommend patterns that conflict with existing config
  • Be vague ("consider improving") - be specific
  • Force-fit patterns that don't apply
  • Overwhelm with too many suggestions
</constraints>