Borrowing it
Nothing to install: this file belongs to tsuru0805/llm-nursery. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/tsuru0805/llm-nursery/main/AGENTS.mdgit clone --depth 1 https://github.com/tsuru0805/llm-nurseryWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/tsuru0805/llm-nursery/agents-md)<a href="https://agentmods.dev/instructions/tsuru0805/llm-nursery/agents-md"><img src="https://agentmods.dev/badge/instructions/tsuru0805/llm-nursery/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.04881 | $0.04881 |
| Opus 5 | $0.02440 | $0.02440 |
| Sonnet 5 | $0.00976 | $0.00976 |
| Haiku 4.5 | $0.00488 | $0.00488 |
Grade A, and why
llm-nursery AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 128 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md · 机看版(给 AI 编码代理 / 二次开发者)
人看版(概念/玩法/快速开始)在 README.md。本文件只放结构化工程事实: 模块职责、数据模型、不变量、env、扩展点、禁改红线。改代码前先读「不变量」与「禁改」两节。
一句话架构
driver(短命子进程 CLI)→ child(生命周期+事务)→ model/decoder/guard(大脑三层)
→ db(SQLite 单库/caregiver)→ scheduler(tick:夜奶/偷学/睡眠整理/观察/outbox)→ events(里程碑/结局)
→ psyche(可选 LLM 心理层,fail-open)+ chunks/bond/observer/portrait(v0.2 养成取舍面)
→ asks/choices/chains/friction/magic/sickness/visible_growth(v0.3 中后期玩法面,全部幂等 fail-open 挂 tick)
→ letters/letter_prompts(v0.4 成年书信:毕业后 tick 只走信箱)。接入面:toolface(白名单,纯标准库)/ server.py(MCP 壳)。
平台:文件锁用 fcntl,POSIX only(macOS/Linux);本地时区口径,无 DST 处理(已知欠账)。
模块表
| 文件 | 职责 | 关键入口 |
|---|---|---|
| nursery/model.py | 插值式可变阶字符 Markov(0..5 阶同时计数);快照 zlib+json+sha256 | VariableOrderMarkov.feed/sample_next/to_blob/from_blob |
| nursery/decoder.py | 阶段化解码;唯一说话出口 | speak(model, guard, stage, rng, ...) -> SpeakResult |
| nursery/guard.py | 反复读护栏(LCS 6-gram 倒排预筛 + 4-gram shingle 比率)+ PII 遮盖 | OverlapGuard.add_source/check, scrub_pii |
| nursery/child.py | 出生/孵化/命名/阶段推导/状态机(惰性结算)/喂语料/说话/找回;全部事务纪律在此 | create_child/hatch_child/pick_name/name_babble/feed_corpus/child_speak/apply_action/ChildBrain |
| nursery/db.py | schema v12 + 事务化迁移(旧档打开即升级;v9 起同时钉 per-child rules_v3_since 戳;v12=letters 表);连接 PRAGMA(WAL/FK/busy_timeout) | connect(db_path)(db_path 必填,无默认) |
| nursery/events.py | 里程碑(含首次四连)/每日确定性抽签(送礼藏品卡)/语出惊人(滚动 7 天配额+锚滤渣)/夜哭忽视/出走/毕业过渡弧+告别窗+五结局(v0.4:预告→他提出离开→3 天窗→窗满自告别)/阶段装订/生日/逐夜账 | tick_events / tick_farewell_arc / farewell_window / in_departure_window / judge_ending / closed_cry_nights |
| nursery/asks.py | 需求事件:当日确定性排班/到点真开口/窗关记账(接住=响应者进账,漏窗不惩罚)+告状稿派生(action_log 真账) | plan_asks/fire_due_asks/settle_asks |
| nursery/choices.py | 选择题两难:swear 触发+抽签模板(每模板一生一次)/choose 拍板(谁先拍算谁的)/超时自决;词块偏置+拍板语料 | plan_choices/fire_due_choices/resolve_choice/settle_choices |
| nursery/chains.py | 连续剧事件链:3 天一条,集间「照护人是否介入」(actor != system)定末集分支;断更废弃 | plan_chains/fire_due_chain_eps |
| nursery/friction.py | 摩擦轴:被晾(observer 公共口径)/摔门/深夜彩蛋/顶嘴取词;唠叨/台阶账在 child 侧同事务 | tick_friction/annoy_stage/recent_direct_anchors |
| nursery/magic.py | 真实语料魔法:时空穿越(真日期)/误译/故事复述/送礼说话——全部真语料派生,查不出不发 | tick_magic |
| nursery/sickness.py | 生病 arc:低频开窗/病中夜叫(夜哭同形)/痊愈;病窗解码扰动挂 child_speak 热路径(indexed 查询) | tick_sickness/open_sickness_date |
| nursery/visible_growth.py | 可见成长:宝贝盒(词块 top 派生)/小本子(心理决策锚词,绝不贴原文)/生日 set-piece | tick_growth/treasure_list |
| nursery/chunks.py | 家庭词块索引:提取(子串吸收)/场景加权抽选/睡眠整理重建(派生数据,可随时全量重建) | consolidate_daily/pick_chunk/rebuild_index |
| nursery/observer.py | 观察日志:晚间从真实统计派生旁观行,查不出=不发 | daily_observe |
| nursery/bond.py | 双照护人关系四维(亲近/安心/踏实/委屈)分账;历史半额估底 | apply_locked/read_bond/bond_trends |
| nursery/portrait.py | 成长画像:全量纯派生 JSON(零写入,不打标签) | build_portrait |
| nursery/scheduler.py | tick 编排:夜奶排班→到期触发→events(含开窗)→窗口静默闸→asks/choices/chains→偷学→魔法/生病→outbox 投递;graduated 分支只走信箱(letters+settle 悬窗);心理层挂尾 | tick_all()/tick_one(db_path, viewer) |
| nursery/psyche.py | 三轴规则表(程序层)+ LLM 结构化决策(fail-open)+ 锚词接力 | apply_rules_locked/maybe_decide/latest_anchor_words |
| nursery/letters.py | v0.4 成年书信:来信排程(结局基调+回信提前钳未来)/信体生成(LLM 骨架或无 key 本地模板,失败顺延+降频)/写信/探望两段/信箱读口(未读/读某封/分页) | tick_letters_fast/deliver_due_letters/write_letter/tick_visit/in_visit/mailbox_summary/read_one_letter/graduation_portrait |
| nursery/letter_prompts.py | 五结局信件骨架+通用生成模板(文案层,可整体替换;占位符名不能改) | 常量 |
| nursery/sampler.py | 偷学抽样;外部存档只读硬闸;schema 约定见文件头注 | connect_archive/sample_fragments |
| nursery/driver.py | CLI:persona 指令/mama 通道(JSON)/--tick/--init-birth;flock | run/dispatch/init_birth/main |
| nursery/toolface.py | 工具面白名单(玩法指令 only,运维入口不放行),零 mcp 依赖 | nursery(player, command) |
| nursery/texts.py / psyche_prompt.py | 文案层(全部玩家文案:出生/命名/照料/事件/夜哭/出走)/ 心理 prompt——可整体替换,逻辑零改动 | 常量 |
| nursery/config.py | 大部分数值:阶段表/解码参数/动作效果/黑暗值/心理轴/预算闸(调度窗口在 scheduler,护栏阈值在 guard,试错上限在 decoder) | 常量 |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago Changed · +6 lines · +664 tokens per session 50dd917893fe
- 8d ago First seen · 122 lines · 4,217 tokens per session scan A 11543beb3712
llm-nursery AGENTS.md is an instructions file published in the GitHub repository tsuru0805/llm-nursery (34 stars, last pushed 6d ago), licensed MIT. It adds 4,881 tokens to every session, about $0.0244 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.