name	project-memory-loop
owner	sophomores
description	Use when non-trivial project work must read project memory before execution and may need stable write-back after. 当非琐碎项目任务必须先读项目记忆, 且收尾时可能需要稳定写回时使用.

Project Memory Loop

Overview

这个 skill 用于长生命周期仓库, 让稳定记忆在执行前生效, 在收尾时做可信写回.
.codex/memory/ 是 canonical memory, 不是 task-state 存储层.
runtime 默认先 consult 项目 memory, 再 consult ~/.codex/memory/ 作为个人全局 overlay.
memory 可以充当 skill 的轻入口:
- canonical memory 保存长期知识与 route 词.
- thin entry memory 把高频旧入口导回既有 skill.
默认先收敛到已有 skill, 不为单一 runbook 新建 skill.
默认走快路径; 治理, promotion, hygiene, verifier 等流程按需触发.

Always Read

rules/core-boundaries.md
workflows/runtime-protocol.md

Common Tasks

为仓库补齐 .codex 项目层 -> 按 workflows/runtime-protocol.md#bootstrap-when-missing
在记忆治理仓库中启动非琐碎任务 -> 按 workflows/runtime-protocol.md#start-of-non-trivial-work
收尾并判断是否写回长期 memory -> 按 workflows/closeout-protocol.md
做周期性治理或 promotion 判断 -> 读 workflows/memory-hygiene-and-enrichment-loop.md 或 workflows/memory-skill-promotion-loop.md
更新 handbook 或面向人的仓库文档 -> 读 references/handbook-update-rules.md 和 references/handbook-style.md
Other / unlisted task -> 先完成 Always Read, 只有需要治理细节时再读 references/memory-governance.md

Known Gotchas

默认启动只要求 ov -> 读 context.md / must_read -> 执行; r 只在需要命中既有 runbook 或 asset 时再运行.
先读 registry.md 或 index.md 会削弱 runtime 路由质量 -> 见 references/known-gotchas.md#route-skipping
task-state 不属于 .codex/memory/ -> 见 references/known-gotchas.md#task-state-pollution
evolution sidecar 不能绕过 canonical memory flush -> 见 references/known-gotchas.md#sidecar-boundary-drift
home memory 是 runtime overlay, 不是项目 memory 的默认写回目标 -> 见 rules/core-boundaries.md

References

Core

rules/core-boundaries.md
workflows/runtime-protocol.md
workflows/closeout-protocol.md

On-demand

references/known-gotchas.md
references/memory-governance.md
references/gpt54-agent-loop.md
references/handbook-update-rules.md
references/handbook-style.md
workflows/memory-hygiene-and-enrichment-loop.md
workflows/memory-skill-promotion-loop.md
references/项目说明书模板.md
references/项目说明书.reference模板.md
scripts/codex_memo.py
scripts/bootstrap_project_codex.py
scripts/memory_tool.py
scripts/build_asset_index.py

Project Memory Loop

Overview

这个 skill 用于长生命周期仓库, 让稳定记忆在执行前生效, 在收尾时做可信写回.

.codex/memory/ 是 canonical memory, 不是 task-state 存储层.

runtime 默认先 consult 项目 memory, 再 consult ~/.codex/memory/ 作为个人全局 overlay.

memory 可以充当 skill 的轻入口:

canonical memory 保存长期知识与 route 词.
thin entry memory 把高频旧入口导回既有 skill.

默认先收敛到已有 skill, 不为单一 runbook 新建 skill.

默认走快路径; 治理, promotion, hygiene, verifier 等流程按需触发.

Always Read

rules/core-boundaries.md

workflows/runtime-protocol.md

Common Tasks

为仓库补齐 .codex 项目层 -> 按 workflows/runtime-protocol.md#bootstrap-when-missing

在记忆治理仓库中启动非琐碎任务 -> 按 workflows/runtime-protocol.md#start-of-non-trivial-work

收尾并判断是否写回长期 memory -> 按 workflows/closeout-protocol.md

做周期性治理或 promotion 判断 -> 读 workflows/memory-hygiene-and-enrichment-loop.md 或 workflows/memory-skill-promotion-loop.md

更新 handbook 或面向人的仓库文档 -> 读 references/handbook-update-rules.md 和 references/handbook-style.md

Other / unlisted task -> 先完成 Always Read, 只有需要治理细节时再读 references/memory-governance.md

Known Gotchas

默认启动只要求 ov -> 读 context.md / must_read -> 执行; r 只在需要命中既有 runbook 或 asset 时再运行.

先读 registry.md 或 index.md 会削弱 runtime 路由质量 -> 见 references/known-gotchas.md#route-skipping

task-state 不属于 .codex/memory/ -> 见 references/known-gotchas.md#task-state-pollution

evolution sidecar 不能绕过 canonical memory flush -> 见 references/known-gotchas.md#sidecar-boundary-drift

home memory 是 runtime overlay, 不是项目 memory 的默认写回目标 -> 见 rules/core-boundaries.md

References

Core

rules/core-boundaries.md

workflows/runtime-protocol.md

workflows/closeout-protocol.md

On-demand

references/known-gotchas.md

references/memory-governance.md

references/gpt54-agent-loop.md

references/handbook-update-rules.md

references/handbook-style.md

workflows/memory-hygiene-and-enrichment-loop.md

workflows/memory-skill-promotion-loop.md

references/项目说明书模板.md

references/项目说明书.reference模板.md

scripts/codex_memo.py

scripts/bootstrap_project_codex.py

scripts/memory_tool.py

scripts/build_asset_index.py

project-memory-loop