sync-comparisons - SKILL.md Agent Skill

name: sync-comparisons description: Sync cross-agent comparison (Dimension 8) across all analysis docs

Sync Comparisons

Update Dimension 8 (Cross-Agent Comparison) sections across all analysis docs so every analyzed agent references every other analyzed agent.

Trigger

/sync-comparisons — sync all analyzed agents' Dimension 8
/sync-comparisons <agent> — only update the specified agent's Dimension 8

Prerequisites

At least 2 analyzed agents must exist (docs/*.md excluding TEMPLATE.md).

Workflow

1. Collect Analyzed Agents

List all docs/*.md files (exclude TEMPLATE.md). Each filename docs/<name>.md corresponds to an analyzed agent. Verify the file has substantive content (not just an empty template).

2. Read Core Characteristics

For each analyzed agent's doc, extract key traits from these dimensions:

Source Dimension	What to Extract
Dimension 1 (Overview)	语言、架构定位、项目类型
Dimension 2 (Agent Loop)	循环模式（单轮/多轮/事件驱动）
Dimension 3 (Tool 系统)	工具注册/调用机制
Dimension 5 (Context 策略)	上下文管理方式
Dimension 6 (错误处理)	错误恢复策略
Dimension 7 (关键创新)	核心特色/独特设计

3. Check Coverage

For each agent (or the specified agent), check its Dimension 8 section (## 8. to next ## or EOF):

Which other analyzed agents are already referenced?
Which are missing?

If all agents are already covered, report "already up to date" and skip.

4. Update Missing References

For each agent with missing references in Dimension 8:

If Dimension 8 only has generic comparison (e.g. "vs 通用 Agent 模式" without referencing specific agents): replace with concrete comparisons against all analyzed agents
If Dimension 8 already has some agent comparisons but is missing others: add the missing agents to the existing comparison structure
Preserve existing content: do not modify or remove existing comparison text, only add missing agent columns/sections

5. Comparison Format

Use a multi-column comparison table when the number of agents (including self) is 4 or fewer:

| 维度 | agent-a | agent-b | agent-c |
|------|---------|---------|---------|
| 语言 | Python | Rust | TypeScript |
| Agent Loop | ... | ... | ... |
| Tool 系统 | ... | ... | ... |
| Context 策略 | ... | ... | ... |
| 编辑方式 | ... | ... | ... |
| 安全/错误处理 | ... | ... | ... |
| 扩展性 | ... | ... | ... |
| LLM 支持 | ... | ... | ... |

When agents exceed 4, use multiple pairwise "vs" sections instead:

### vs agent-b

（对比要点...）

### vs agent-c

（对比要点...）

6. Update Summary

After the comparison table/sections, update or add a summary paragraph that:

Highlights the most significant differences
Notes complementary strengths
Covers all compared agents (not just the newly added ones)

Comparison Dimensions

The comparison should cover these aspects (as table rows or discussion points):

维度	说明
语言	实现语言和技术栈
Agent Loop	循环架构（单轮请求/多轮对话/事件驱动/ReAct）
Tool 系统	工具定义、注册、调用方式
Context 策略	上下文窗口管理（截断/压缩/摘要/repomap）
编辑方式	代码编辑策略（search-replace/diff/whole-file/patch）
安全/错误处理	权限控制、沙箱、错误恢复
扩展性	插件/扩展机制
LLM 支持	支持的模型和切换方式

Output Requirements

Write in Chinese, code identifiers in English (matching project conventions)
Focus on differences, not feature lists — describe how each agent's approach differs
Reference docs/codex-cli.md Dimension 8 as a style reference for good comparison tables
Each comparison point should be concise (1-2 sentences per cell)

After Sync

Run npm run lint to verify check #6 (跨 Agent 对比覆盖) passes
No need to update agents.yaml status — this is a doc maintenance task