Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/windameister/oneshard/claude-mdgit clone --depth 1 https://github.com/windameister/oneshardWrote 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/windameister/oneshard/claude-md)<a href="https://agentmods.dev/instructions/windameister/oneshard/claude-md"><img src="https://agentmods.dev/badge/instructions/windameister/oneshard/claude-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 | $0.01232 | $0.01232 |
| Opus 5 | $0.00616 | $0.00616 |
| Sonnet 5 | $0.00246 | $0.00246 |
| Haiku 4.5 | $0.00123 | $0.00123 |
Grade A, and why
oneshard CLAUDE.md scanned grade A with 1 finding 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 5d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
冒烟(curl 端点 + 前端改动必须 Playwright 实测生产页面真实行为,包括 zh/en 双语态)。 How it starts
The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.
深空沙盒 · 项目约定
对人类和 AI agent 同等可读可玩的持久宇宙 4X 沙盒。当前处于 Phase 0(headless 情绪验证)。
常用命令
npm test # vitest 全量
npm run typecheck # tsc --noEmit
npm run create-universe -- --name alpha # 创世(--seed --systems --civs --tick-seconds)
npm run server -- --data data/alpha # 起服务(--port 8420 --tick-ms 快进)
npm run cli -- --token dsx_xxx # 人类终端客户端
铁律(改代码前读)
- 契约先行:MCP 工具的名字/入参/返回形状改动,必须先改
docs/api-contract.md,再改src/server/mcp.ts与 CLI。 - 无作弊视野:任何给玩家的数据必须经
src/views/views.ts脱敏。永不输出:他人 token/配额/收件箱、迷雾外实时信息。新增字段时先想"这会不会泄露迷雾"。 - 确定性:引擎(src/engine、src/core)禁止
Date.now()/Math.random()——随机走Rng.forState(state)(游标持久化),时间只有state.tick。墙钟只允许出现在 src/server 调度层。 - 抛错=零副作用:行动函数(src/actions)必须先完成全部校验再
spendQuota再变更;抛GameError时不允许有任何状态改动。 - 变更段禁止 await:并发安全靠 Node 单线程 + 同步变更段(见
GameRuntime.runCivAction)。在行动/tick 的变更路径里引入 await 会制造竞态。 - 平衡数值全部住在
src/core/constants.ts,标注均为临时值;改平衡不要动公式所在的引擎代码。 - 设计冲突时回到四支柱裁决(
docs/vision.md):同规则同接口 / 决策质量>行动频率 / 事实层可欺骗物理层不可违反 / 渲染只是另一个客户端。 - 测试先行(TDD,2026-07-18 起):任何 feature/bugfix 按 红→绿→部署验证 三段走: ①先写会失败的测试(引擎/服务端=vitest;数值平衡=balance-sim 矩阵基线;前端=本地起服 Playwright 断言关键行为),确认红;②实现到绿(typecheck + vitest 全量);③部署后生产 冒烟(curl 端点 + 前端改动必须 Playwright 实测生产页面真实行为,包括 zh/en 双语态)。 线上缺陷修复必须附带一条修复前会失败的回归测试(例:tests/auth.test.ts 静态缓存头)。 子 agent 交付不算完成——lead 必须复验可见行为后才可提交;仅 node --check/文本 grep 不构成前端验收(2026-07-18 编年史缓存脏读事故的教训)。
子 agent 分工(Claude 任 lead 总负责,汇总与最终交付)
开发、文档、code review 三类工作都应充分委派给子 agent 并行执行:
- codex(gpt-5.6-sol,已是 ~/.codex 默认):UI/视觉/前端为主,亦可承接一般开发与 code review。
codex exec --sandbox workspace-write -C <repo> --color never - < /tmp/brief.md - GLM 5.2(opencode coding plan):文档/设计文集为主,亦可承接开发与 review。
opencode run -m zai-coding-plan/glm-5.2 "$(cat /tmp/brief.md)"
协作约定:brief 落盘 /tmp/*.md(必读文件清单+交付物+硬约束);每个 agent 划互不相交的
文件范围;共享契约(docs/api-contract.md)由 lead 先行改定,子 agent 只读;子 agent 禁跑
git;两者都后台并行跑(run_in_background),lead 验收集成后统一提交。
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.
- 5d ago First seen · 56 lines · 1,232 tokens per session scan A 4ee7ef8cd2d6
oneshard CLAUDE.md is an instructions file published in the GitHub repository windameister/oneshard (0 stars, last pushed 1mo ago), licensed MIT. It adds 1,232 tokens to every session, about $0.0062 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
vibedgames CLAUDE.md
Claude Code instructions for kyh/vibedgames, covering agent instructions, agent-driven development, what this is, product & business context and key architectural decisions.
vibedgames AGENTS.md
AGENTS.md instructions for kyh/vibedgames, covering agents.md, quickstart (headless), environment, seeded logins and a real session cookie, for handing to a browser context.
promptasy CLAUDE.md
Claude Code instructions for romanticamaj/promptasy, covering promptasy, 這份文件是什麼, /goal — 長期、開放、不設限方向的目標, 遊戲願景與核心體驗 and 設計支柱(design pillars).
promptasy AGENTS.md
AGENTS.md instructions for romanticamaj/promptasy, covering agents.md — promptasy 工程 harness, 專案一句話, 每個 phase 的節奏, 測試金字塔(由快到慢) and 測試策略(成本控制的核心).
Legends-Of-Heroes AGENTS.md
Instructions for FlameskyDexive/Legends-Of-Heroes, covering agents.md, 最小入口规则, 两套技能体系分工, unity 交互入口优先级 and aibridge bootstrap.
advjs CLAUDE.md
Instructions for YunYouJun/advjs, covering claude.md, project overview, common commands, run a single unit test file and dev specific packages.