Incremental Build

Incremental Build / 增量验证与接口变更协议

> Batch edits hide each other's errors. Verify after every file so a failure points at one change. > 批量编辑会让错误互相掩盖。每个文件后都验证,失败才能指向单一变更。

Incremental build / 增量编译验证

1. After each file edit, immediately run the matching build/check — compiled langs: compile; TS: tsc --noEmit; Python: ruff/mypy or import the module; no type-checker: run the smallest test covering that file.

每编辑完一个文件立即运行对应 build/检查——编译型:编译;TS:tsc --noEmit;Python:ruff/mypy 或 import 该模块;无类型检查:至少跑覆盖该文件的最小单测。

1. Never edit 3+ files before the first build — masked errors compound, fix cost grows exponentially.

禁止连续编辑 3+ 文件后才首次 build——被掩盖的错误叠加,修复成本指数增长。

1. Agent sub-tasks must require "build after each file edit" in the task description.

派给 agent 的任务描述里要写明"每文件编辑后 build 验证"。(派子任务的完整必含项见 agent-team 的「派子任务必含清单」)

Interface-change protocol / 接口变更协议

1. Search: grep every type implementing the interface (including mocks / test doubles).

先搜:grep 所有实现该接口的类型(含 mock / test double)。

1. List: enumerate each file + line that needs the change.

列清单:列出每个要改的文件和行号。

1. Change all at once: interface definition + all implementations in the same edit batch.

一次性全改:接口定义 + 所有实现在同一编辑批次内完成。

1. Verify immediately: build right after, confirm zero compile errors.

立即验证:改完立即 build,确认零编译错误。

When this triggers / 触发时机

• About to edit 3+ files in a row.
• About to modify an interface / trait / abstract class / protocol with 2+ implementations (counting mocks / test doubles).
• About to hand an agent a multi-file change.
• You've edited a batch and haven't built yet (stop and build now).

Done criterion (verifiable) / 完成判据（可验证）

✅ Each edited file was built/checked before the next one; for interface changes, every implementation found by grep was updated in one batch and the build is green. ⚠️ A file was edited but the build/check couldn't run (no compiler, env not set up, etc.) — say so explicitly and treat it as unverified, don't move on as if it's green. ⚠️ Some files built, but not every grep'd interface implementation was updated yet — list the implementations still unverified. ❌ Several files edited, no build yet — pause and verify before continuing.

Worked examples / 实战反例

Real before/after cases for this discipline live in [EXAMPLES.md](./EXAMPLES.md) — read them before you act. 本纪律的真实 before/after 反例见 [EXAMPLES.md](./EXAMPLES.md) —— 动手前先对照。

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: yli769227-jpg
• Source: yli769227-jpg/agent-discipline
• License: MIT

Install and usage instructions live in the source repository linked above.