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/superdeman/cockpit-agent/agents-mdgit clone --depth 1 https://github.com/SuperdeMan/cockpit-agentWhat 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 | $0.03690 | $0.03690 |
| Opus 5 | $0.01845 | $0.01845 |
| Sonnet 5 | $0.00738 | $0.00738 |
| Haiku 4.5 | $0.00369 | $0.00369 |
Grade A, and why
cockpit-agent 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 2d 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 — 237 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — 接手者入口
先读本文件，再动代码。工程规则最高权威是
CLAUDE.md；架构唯一真相源是docs/architecture/cockpit-agent-architecture.md。 当前 QA 与发布交接统一看docs/reviews/2026-08-30-qa-closeout-handoff.md。 逐批历史只查docs/agents-history.md，不要把历史流水抄回本文件。
1. 项目是什么
云边协同的智能座舱 multi-agent 系统。端侧快系统处理高频、安全敏感和离线能力；云侧 Planner 处理复杂、多域、多轮任务。Agent 统一使用 gRPC 契约 + Manifest，经 Registry 发现； 车控只经 VAL，LLM 只产意图/计划。
当前阶段是 Phase 1 工程化 PoC。工程主干、云端中枢、真实 Provider、语音回路、记忆、 可观测、旅程验证、M0a→M4、M5 数据飞轮与探索式 QA 的编号开发批均已落地。PoC 已可运行， 但真实 CAN/SOME-IP、量产账号体系、完整隐私治理和部分外部能力仍是明确边界。
2. 文档地图
| 想了解 | 权威入口 |
|---|---|
| 当前 release、测试证据、QA 活项 | docs/reviews/2026-08-30-qa-closeout-handoff.md |
| 工程规则、目录、安全红线 | CLAUDE.md |
| 全局架构 | docs/architecture/cockpit-agent-architecture.md |
| Phase 1 计划与量产 DoD | docs/architecture/phase1-implementation-plan.md |
| 环境、端口、命名、错误码 | docs/conventions.md |
| 本地/云端开发与部署 | docs/dev-guide.md |
| 测试分层与 E2E | test/README.md |
| 意图对抗测试 | docs/guides/intent-adversarial-testing.md |
| 真实 Provider 接入 | docs/guides/provider-integration.md |
| MiniMax 原始 QA 问题 | docs/reviews/2026-08-26-minimax-cloud-qa-findings.md |
| MiniMax 根因与修复批 | docs/design/2026-08-27-minimax-qa-root-cause-fix-plan.md |
| 安全确认写闸 | docs/design/2026-08-30-qa-safety-confirmed-write-guard.md |
| Android App 当前计划 | docs/design/2026-08-30-mobile-ux-v2-b2-implementation-plan.md |
| 历史流水 | docs/agents-history.md（只追加） |
服务子目录各有 README；改某个服务前先读该目录 README。
3. 不可违反的规则
3.1 运行环境
- 根目录
.env是唯一运行时环境与密钥来源；不得复制、维护或依赖deploy/.env。 - 本地真栈只用
make up或根compose.yaml；不得以deploy/docker-compose.yaml为首文件。 - 任何 E2E、Compose、部署或脚本/manifest 改动前，先从仓库根读取
dev-stack.local。 dev-stack.local只允许target=local|cloud，不得保存 token、密码、私钥或 URL。target=cloud时禁止启动本地 Compose；本地只做编辑、单测、静态检查和 Vite。- 不停止其他 agent 正在使用的 Docker、Metro、Gradle、pytest 或真栈进程。
3.2 云端发布
- 真栈动作前先运行
python scripts/dev_stack.py target show。 - Windows 真栈、SSH 与 deploy 命令一律用 PowerShell；Git Bash/MSYS 会改写参数引号。
- cloud deploy 只接受 clean、已提交、main 可达的 SHA；先 dry-run，再经人工授权
--apply。 - deploy 不自动 commit、merge 或 push；
git push必须单独授权。 - push 前必须逐条展示
origin/main..HEAD，让用户看到会被一并推走的提交。 - 未标
remote_safe的 E2E 不在 cloud 缺省运行；remote_mutating=true仍需精确--id、--allow-mutating和本轮人工授权。 - 支付、商户写、真实车控、数据删除、系统配置不因通用 deploy/E2E 授权自动放行。
- 三存储迁云只用
scripts/cloud_data_migration.py；final 必须先取得停写授权。 - 不修改
.env、安全组、Tailscale、CI/CD、systemd、数据库 schema，除非用户逐项授权。
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.
- 2d ago First seen · 237 lines · 3,690 tokens per session scan A 784cfb6b1b31
cockpit-agent AGENTS.md is an instructions file published in the GitHub repository SuperdeMan/cockpit-agent (14 stars, last pushed 2d ago), licensed Apache-2.0. It adds 3,690 tokens to every session, about $0.0185 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
agentic-tools AGENTS.md
Instructions for meta-quest/agentic-tools, covering agent guide — meta quest agentic tools, what this repo is, how skills work, shared references and skill index.
open-xiaoai-bridge AGENTS.md
Instructions for coderzc/open-xiaoai-bridge, covering agents.md, 系统架构, 项目结构, 核心组件 and mainapp (core/app.py).
agenticros AGENTS.md
Instructions for agenticros/agenticros, covering agenticros, what this is, important: controlling the robot from codex, architecture and key source files.
NextE AGENTS.md
Instructions for erosTeam/NextE, covering nexte agent entry, before acting, hard stops, state management v2 and device operations.
hass-mcp-server AGENTS.md
Instructions for ganhammar/hass-mcp-server, covering working on this integration, calling into home assistant, before replacing a private call with a service, anything left in category 2 or 3 and conventions.
orchestrator AGENTS.md
Instructions for c9r-io/orchestrator, covering agent orchestrator — ai dev platform index, project overview, core capabilities, architecture and manifest format.