菜单
Living project memory files: .agents/PROGRESS.md, .agents/LESSONS.md, docs/decisions.md, session-log, agent-notes. Use when finishing a task/session, writing handoff, night audit, onboarding a cold project, or user says progress/lessons/ADR/handoff.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/VKirill/claude-lane-stack --skill project-memory命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Solo file-based multi-lane orchestration. Durable switchable Kimi/Qwen/AGY/Grok controller, no daytime LLM review, nightly Codex review/fix, PM auto-merges to main. Use when planning runs, splitting DAGs, dispatching writers, recovering blocked tasks, or shipping to main.
Primary project onboarding for Claude Lane Stack. Diátaxis-classified LLM pack (MODULE_MAP, API_SURFACE, TEST_INDEX, TAXONOMY, RUNBOOK, Google DESIGN.md). Schemas + validation in references/. Weekly docs-maintainer. Use when: /project-onboard, онбординг, init project, bootstrap CLAUDE.md, deep onboard, wiki→llm docs.
OpenAI Codex CLI — OpenAI's official Rust-based agentic terminal coding tool (NOT the deprecated 2021 Codex completion model). Use when: openai codex, codex CLI, @openai/codex npm, codex exec, codex exec review, codex app-server, AGENTS.md, .codex/config.toml, sandbox_mode read-only/workspace-write/danger-full-access, approval_policy never, service_tier fast, gpt-5.6-luna/terra/sol, lane-writer profile. SKIP: Claude Code CLI (→claude-code), OpenCode CLI (→opencode), OpenAI SDK programmatic (→openai-sdk), GitHub Copilot CLI (different tool), deprecated code-davinci-002 model (discontinued).
基于 SOC 职业分类
正在显示 SKILL.md
| name | project-memory |
| description | Living project memory files: .agents/PROGRESS.md, .agents/LESSONS.md, docs/decisions.md, session-log, agent-notes. Use when finishing a task/session, writing handoff, night audit, onboarding a cold project, or user says progress/lessons/ADR/handoff. |
Canon: /home/ubuntu/.agents/docs/PROJECT-MEMORY.md
Templates: ~/.agents/templates/
Init: ~/.agents/bin/project-memory-init <repo>
| File | Max size | Update when |
|---|---|---|
.agents/PROGRESS.md | ~40 lines | End of meaningful work / session |
.agents/LESSONS.md | grow slowly | After user correction or failed approach |
docs/decisions.md | rare | Expensive irreversible choice |
.agents/session-log/* | auto | Hooks (do not hand-author bulk) |
.agents/agent-notes/OPEN.md | grow | Debt / simplify later |
Canonical location is under .agents/ so root cleanups do not wipe living state.
If only root PROGRESS.md / LESSONS.md exist, project-memory-init and session-ledger migrate them into .agents/.
AGENTS.md / CLAUDE.md (project)..agents/PROGRESS.md if present (one pass; fallback: root PROGRESS.md)..agents/LESSONS.md titles if present..agents/todos/INDEX.md counts only.When report.md is green / complete:
.agents/PROGRESS.md — rewrite Now / Next / Last verify (keep Auto section intact)..agents/LESSONS.md one entry.agent-notes/OPEN.md.## Now
- <one sentence current reality>
## Blocked
- <or "none">
## Next
- [ ] <next concrete step>
## Last verify
- command: <what you ran>
- result: green|red
- when: YYYY-MM-DD
Leave <!-- auto:session-ledger --> … <!-- /auto:session-ledger --> alone (hooks own it).
### YYYY-MM-DD — short title
- **Symptom:** …
- **Wrong approach:** …
- **Do:** …
- **Don't:** …
- **Evidence:** path or test name
## ADR-NNN: Title
- **Date:** YYYY-MM-DD
- **Status:** accepted
- **Context:** …
- **Decision:** …
- **Consequences:** …
- **Alternatives considered:** …
agent-todos, not PROGRESS.progress_now, close_next, and
close_open actions in run.yaml; post-merge run-finalize applies them and
records hashes/actions in finalize.json. Never guess stale checklist items.Run ~/.agents/bin/night-audit . or read AUDIT-*.md. Close OPEN items or spawn todos/runs.
.agents/)