ECC
ECC 对比与选型：为团队选择合适的代理 Harness
比较 ECC、原生客户端插件、轻量 Prompt 包与自定义编排在控制、兼容性和运营成本上的差异。
你将学会
- Compare agent control planes
- Run a fair harness bake-off
- Document selection and fallback criteria
开始前需要
- Basic Git and command-line usage
- Comfort reading a project README
You can explain the project, run its documented first step, and decide what to verify before adopting it.
先看结论
- Compare host parity, hooks, memory, permissions, observability, and uninstall—not catalog size.
- ECC suits repeatable cross-client loops when the team can govern generated state.
- Use identical tasks and safety gates in a bake-off, then keep a manual fallback.
Compare control planes
选型应比较宿主兼容性、Hook 语义、权限默认值、持久化状态、可观测性、升级和卸载，而不是把共享文件名当成行为一致性的证明。
Use control-plane criteria rather than feature counts: host support, hook semantics, permission defaults, persisted state, observability, upgrade path, and uninstall completeness. Shared filenames do not prove behavioral parity.
When each option fits
Choose ECC when a team wants a repeatable plan-test-review loop across several supported clients and is prepared to govern generated files and memory. Choose a native plugin when one host's events and UI integration are more important than portability.
Choose a lightweight prompt pack for low-risk, mostly manual workflows. Choose a custom orchestrator when compliance or domain-specific events require a smaller, fully owned surface and the team can maintain it.
Run a fair bake-off
Use the same harmless repository task, model, shell, and approval policy. Compare accepted-change rate, test coverage, context tokens, hook/MCP latency, memory writes, blocked-command behavior, adapter gaps, and operator review time. Preserve logs and generated manifests for every candidate.
A faster harness that omits an approval or redaction boundary is not an equivalent win. Score safety and explainability as release gates, not optional features.
Make and revisit the decision
Adopt ECC when its catalog and cross-client workflow reduce repeated engineering work without exceeding the team's review capacity. Keep a documented fallback to the native client or manual checklist, and test uninstall before expanding permissions.
Revisit the choice when the model, editor, repository permissions, or compliance requirements change. A versioned matrix prevents a future upgrade from silently changing the accepted operating boundary.
如何选择
| 比较维度 | 方案 A | 方案 B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
实施步骤
- 1
Define control, safety, parity, and operating-cost criteria.
- 2
Run ECC, a native plugin, and a lightweight/custom alternative on one fixed task.
- 3
Compare quality, context, hook/MCP latency, memory, safety, and review effort.
- 4
Version the decision matrix, fallback, and uninstall test.
可复制示例
candidate | host parity | hook control | memory scope | observability | uninstall | review cost
ECC | matrix | explicit | scoped | event trace | test | measure
native plugin | host-specific | native | host policy | host tools | verify | measure常见问题
Is ECC automatically the best choice for every team?
No. Its breadth helps only when the team can review and maintain generated hooks, MCP entries, memory, and adapters.
How do I compare a native plugin fairly?
Use the same task, model, permissions, approval policy, and safety gates, then compare accepted changes and operating evidence.
资料来源
- ECC README (captured 2026-08-31)来源核查 2026-08-31
- ECC repository来源核查 2026-08-31