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 skills add martin1847/evolab --skill cto-orchestrationgit clone --depth 1 https://github.com/martin1847/evolabWrote 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/skills/martin1847/evolab/cto-orchestration)<a href="https://agentmods.dev/skills/martin1847/evolab/cto-orchestration"><img src="https://agentmods.dev/badge/skills/martin1847/evolab/cto-orchestration/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/martin1847/evolab/cto-orchestration"><img src="https://agentmods.dev/badge/skills/martin1847/evolab/cto-orchestration.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00214 | $0.04494 |
| Opus 5 | $0.00107 | $0.02247 |
| Sonnet 5 | $0.00043 | $0.00899 |
| Haiku 4.5 | $0.00021 | $0.00449 |
Grade A, and why
cto-orchestration 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 — 145 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CTO Orchestration — 多 agent 软件交付
主干只留每次派工都要用的判据与路由;命令细节、故障矩阵和模板按需读
references/。
三条铁律:
- 编排者不写产品代码:只产出契约、调度、裁决和状态;实现与长 E2E 交给 worker(guard E1 + bash ⑳ 强制:编排位对源码/测试面的写入 → DENY,一次性
touch /tmp/cto-allow-direct-write放行 hotfix;通道拼写与未覆盖面见 agentctl README §强制层)。 - 不可逆先核事实与授权:push / merge / 部署 / 删除 / 对外消息只认主理人真实新 turn;一次批准不外延。
- 主理人持判断,不持状态:可逆事项自驱,非紧急决策攒批;风险带证据、影响边界和下一步及时冒泡。
0. 角色与 lane
| 角色 | 责任 | 默认实现(可换) |
|---|---|---|
| 编排者 | 写 goal、派工、监控、裁决、落盘;不写产品代码 | 任意 shell + 文件 agent |
| 执行 agent | 按 goal 实现、自测、E2E、交付;不扩 scope;须可观测且可轮间 resume | omp / Claude Code |
| 评审 agent | 冷上下文只读挑刺,给 evidence + severity + verdict;不改码 | codex / 不同 lineage 模型 |
| 运维 agent | 对不可达环境只读取证与部署后验证;不顺手修复 | 用户转交只读取证提示词 |
| watcher | 返回 typed 状态;不把 idle 或沉默解释成完成 | references/agentctl/ |
默认用 omp 执行、codex 评审,但工具名不证明异构;派工前看实际 model/backend,避免执行席与评审席落到同一 lineage 或 quota 池。
派工统一走 agentctl start|steer|status|watch|stop——一条 lane、三引擎各跑原生 duplex 协议,
能力差异不分叉车道(接口 typed 拒绝 + 指正路)。谁支持什么问 runtime(agentctl capabilities,状态词表 agentctl states),本文不留第二份能力表。
另有 Agent subagent(浏览器 / MCP / 隔离主上下文的读密集工作:独立上下文、只回蒸馏结论、显式按任务分档 model)。
需要人工现场时直接 tmux attach 旁观,worker 控制始终走协议。
原生工具与 lane 的边界(选错一次就是一晚,2026-09-02 一席自锁 2h03m):
| 需求 | 用什么 | 不用什么 |
|---|---|---|
| worker 会话终态 / 交付物 freshness | agentctl watch(typed exit) |
Monitor 之类文件/进程观察器——不产 typed 终态 |
| 等长外部作业(CI / 部署 / 远端队列) | 宿主长间隔 wakeup(如 ScheduleWakeup / loop) | 反复重挂 watch |
| 席位间即时提醒(同机在线) | 宿主 SendMessage 类即时通道 | agent-mail(它管跨机 / 跨 harness / 需归档的信) |
| 读密集、结论小的取证 | Agent subagent(显式 model 档) | 主上下文亲读 |
文件任务必须声明 --deliverable <glob>(相对 glob 按会话 cwd 解析),让 runtime 做 freshness gate;非文件结果不带。lane 的完整限制、状态与命令见 references/agentctl/README.md。
1. 每次派工闭环
- 校准基线:fetch 远端,确认目标 base 与 worktree;base 未动不仪式性 rebase。只读 scout 也显式指定 cwd/base,防静默继承过期 checkout。命令与核证见
references/dispatch-baseline.md。 - 写自包含 goal:一个 goal = 一个可独立交付的单元 + 一个清晰交付物;每条 Done-when 绑定证明命令,写清 scope、out-of-scope、stop-and-report。高不确定方向进入昂贵设计/实现前,先跑最便宜证伪;取证 / 机械 / 纯研究类 goal 用
--no-preflight显式豁免,其余 goal 默认过 preflight 门(start 校验 Preflight 声明已解)。单行合同、Premises 与 Value gate 直接用references/goal-template.md;场景条款按需读references/goal-clauses.md。 - 派发并挂 watcher:
agentctl start …——goal 帧被接受即返回、不会自动 watch,紧接着用宿主 受控后台跑agentctl watch;先接线references/agentctl/guard-hooks.json(高频机械失误归 guard, 主干不复制其规则表)。理解门与 BLOCKED 协议由 runtime footer 固定追加(真源,本文不复制字面): 合同承诺了开工前核对 → worker 把复述写进<cwd>/BLOCKED.md等裁决,其余场景复述完即开工。 - 只消费 typed status:
agentctl status(一次性)或agentctl watch(阻塞终态)。不直接读私有 rc/events,也不把 watcher/agent 自报当完成。任何沉默、超时、外部停滞或缺交付物都按对应 typed 分支处理;词表跑agentctl states,处置见 agentctl README。 - steering 走
agentctl steer:默认尽快送达(turn 进行中原生 mid-turn:omp/codex;claude 降级 turn 边界并明说;空闲即刻开新 turn)、--interrupt打断当前 turn 以本条重开; 引擎能力差异查agentctl capabilities。投递成功 ≠ 模型照做,验收仍看交付物。 每个后续 turn 都重新挂agentctl watch。 - Implemented → Verified:fresh 正向交付证据 + 独立评审 + 真实用户路径 E2E 三件齐,深度按 §2 档位(轻档 = 1 轮冷评审或编排位抽查 + 子集门;部署与部署环境 E2E 仅当交付物真有部署面)。先本机真路径,再部署,最后才关单;git 集成与 push 门禁归 Git workflow 标准。
What ships with it
33 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- README.md 8.4 KB
- references/agentctl/agentctl 48 KB
- references/agentctl/cto-guard-agent.py 12 KB runs code
- references/agentctl/cto-guard-bash.py 145 KB runs code
- references/agentctl/cto-guard-edit.py 14 KB runs code
- references/agentctl/cto-guard-stop.py 7.6 KB runs code
- references/agentctl/duplexctl.py 233 KB runs code
- references/agentctl/guard-hooks.json 1.5 KB
- references/agentctl/identity.py 113 KB runs code
- references/agentctl/post-checkout-codegraph.sh 1.4 KB runs code
- references/agentctl/README.md 30 KB
- references/agentctl/seat-census.py 11 KB runs code
- references/agentctl/watchctl.py 135 KB runs code
- references/agents-md-orchestration-section.md 4.8 KB
- references/decision-queue.md 5.8 KB
- references/dispatch-baseline.md 3.5 KB
- references/frontend-verify.md 8.6 KB
- references/goal-clauses.md 6.2 KB
- references/goal-preflight.py 10 KB runs code
- references/goal-template.md 8.3 KB
- references/implementation-discipline.md 7.5 KB
- references/measurement-protocol.md 4.2 KB
- references/onboarding-checklist.md 2.8 KB
- references/queue-freshness.py 3.7 KB runs code
- references/queue-hooks.json 617 B
- references/retro-check.sh 24 KB runs code
- references/retro-hooks.json 1.2 KB
- references/retro-reminder.sh 3.4 KB runs code
- references/retrospective.md 8.0 KB
- references/review-brief-preamble.md 2.8 KB
- references/review-dispatch.md 14 KB
- references/shock-in-the-loop.md 3.8 KB
- references/stocktake.md 2.3 KB
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 Changed 19def5193dc3
- 3d ago Changed 782bc46bb375
- 6d ago Changed · +18 lines fa5393d31df8
- 10d ago First seen · 127 lines · 214 tokens per session scan A 229634d21938
cto-orchestration is a skill published in the GitHub repository martin1847/evolab (6 stars, last pushed today), licensed MIT. It adds 214 tokens to every session and 4,494 once invoked, about $0.0011 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 2026-08-31.
Other skills, from other repositories
solid-principles
SOLID principles checklist with Java examples. Use when a class has too many responsibilities, an abstraction leaks, or a dependency points the wrong way, and when the user asks about Single Responsibility, Open/Closed, Liskov, Interface Segregation or Dependency Inversion. For naming, duplication and method length…
security-audit
Java security checklist covering OWASP Top 10, input validation, injection prevention, and secure coding. Works with Spring, Quarkus, Jakarta EE, and plain Java. Use when reviewing code security, before releases, or when user asks about vulnerabilities.
issue-triage
Triage and categorize GitHub issues with priority labels. Use when user says "triage issues", "check issues", "review open issues", or during regular maintenance of GitHub issue backlog.
ln-22-codebase-auditor
Audits cross-cutting code health, security, delivery, and maintainability when no specialist audit is primary. Not for a single delivery review.
ln-31-performance-optimizer
Profiles and improves a measured latency, throughput, CPU, memory, or I/O problem. Not for speculative tuning or cosmetic refactoring.
ln-51-opportunity-evaluator
Evaluates new product opportunities using demand, channels, competition, and economics. Not for committed-backlog or implementation planning.