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 Ancienttwo/repo-harness --skill skillgit clone --depth 1 https://github.com/Ancienttwo/repo-harnessWrote 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/ancienttwo/repo-harness/skill)<a href="https://agentmods.dev/skills/ancienttwo/repo-harness/skill"><img src="https://agentmods.dev/badge/skills/ancienttwo/repo-harness/skill/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/ancienttwo/repo-harness/skill"><img src="https://agentmods.dev/badge/skills/ancienttwo/repo-harness/skill.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 | $0.00052 | $0.01555 |
| Opus 5.5 | $0.00021 | $0.00622 |
| Sonnet 5.5 | $0.00010 | $0.00311 |
| Haiku 4.5 | $0.00005 | $0.00155 |
Grade A, and why
development-scheduler 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 13d 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 — 39 lines — stays where its author put it; the contents beside it link to each section on GitHub.
开发调度
Bot 负责用户对话、目标与验收、派工、监督和结果收集;worker 实现,独立 reviewer 审查。repo-harness session 可直接开发,Bot 模式复用 Host 既有 executor、远控和自动化。第一阶段以现有 task-agent、Herdr、Git 分支与 PR 状态为事实来源,引用原始返回和证据,不另建调度数据库或 Git common-dir runtime store。若将来出现真实并发或交接故障,再另开计划补最小锁。
Kanban 只读展示进度、阻碍、候选和证据;反馈、澄清、改需求、催办和审批跟进回 Bot,经受控工具执行。现有产品写入口尚未移除时,不能宣称 UI 已只读,也不能借 browser write 绕过边界。按需读 Host 接入与 repo-harness 适配。
判断任务是否可派
先读当前要求、项目规则和直接相关计划,确定目标、验收、读写范围、owner、依赖、预算与禁止动作。沿用已有授权直接推进;commit、push、merge、deploy、安装和清理分别受用户授权约束,Skill、建议、PASS 或上传成功都不授予权限。缺失资料或能力时阻塞依赖动作,能独立完成的部分继续,不能合成缺失身份或成功证据。
Herdr 基准版本为 0.9.3;控制前按安装版本读 herdr --skill、herdr --help 和相关命令组帮助,不裸启 Herdr 或用 mutating 命令探路。遵守真实 caller context;外部入口须有明确授权,不伪造 HERDR_ENV。核对目标 machine/server、cwd、live 身份和已有任务,采用返回 ID,不凭 UI 焦点、旧 pane ID 或 idle 判断可复用。真实 provider sessionID 与恢复位置来自可核验的 provider metadata/session 记录,pane、Tab、PID 不能代替它;未知身份阻塞相关派工和清理。
后端默认 Codex(gpt-6.1-sol,high),前端默认 Claude,具体任务限制优先。Claude 保留正常权限审批,不使用跳过权限模式;Claude 只读审查接入另立方案,当前独立审查使用已验证的 Codex 路径。启动任何 CLI worker 都必须复用 Pi 1.0 / OAR,通过现有 task-agent + Herdr 承载;不裸启 provider,不写手工 adapter/parser。安装版本未接通该路径时报告缺口并暂停启动,不以新 adapter 绕过。
按分支目标或独立写边界选择 worktree,避免微任务各建一个;Tab/pane 只是组织视图,不隔离文件。并行 writer 的文件 ownership 不重叠,共有文件交给唯一 owner 或顺序执行。冲突先停写并保存 WIP,前任确认停写后再交接,不能 stash、覆盖或终止他人工作。派工前核对同目标/角色/候选的未完成任务、已有结果和总预算,包含复用进程、descendants 与 reviewer 槽位;容量不足就排队,未知额度不能声称充裕。
按事件跟进,按证据收敛
给 worker 目标、冻结验收、输入来源、owner/读写范围、禁止动作、预算、结果位置和停止条件;记录足够定位任务、真实 session、候选及原始证据的引用,不逐任务抄登记字段表。提交成功不等于执行,须有活动或结果证据。每个派发任务收集结果,或保存明确取消及其确认状态。
按需或在结果、阻碍、审批、停滞、交接等事件发生时检查;长任务使用与风险和耗时相称的有界检查窗口,不默认每 60 秒巡查。quota 耗尽就停新派并保存成果;timeout、失联或 unknown 不证明未执行:先查原任务、活动、候选与 session,未知 writer 不 reset budget、不重复派工。取消未确认仍占预算,不能宣称退出或关闭;诊断无新证据时保存阻塞事实并报告,不无限轮询或偷换 provider。
人工接手先暂停 Bot 新指令与本任务自动发送,收集在途工作,确认 owner 和同一真实 session 后显式交接;Bot 只读观察。当前 owner 明确归还、在途状态与预算/去重核对后才恢复。idle、断连或关窗不是交接凭证,状态不明就保持暂停;不把现有连接或手写记录称作原子 fence。
冻结分支、HEAD/范围与已有候选证据后再审查。独立 reviewer 用未参与设计/实现的 session,只读候选,只写指定 review;advisor 不能改名审自己的设计,跨 provider 也不自动独立。提供原始材料和验收,不附作者自评或预期 verdict。审查结论须绑定同一候选,漂移使旧 PASS 失效。保留 finding、修复依据及复验结果;不能换 reviewer 刷 PASS,也不能把测试通过当独立审查或发布授权。
What ships with it
2 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.
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.
- 13d ago First seen · 39 lines · 52 tokens per session scan A e47ff31617dd
development-scheduler is a skill published in the GitHub repository Ancienttwo/repo-harness (434 stars, last pushed today), licensed MIT. It adds 52 tokens to every session and 1,555 once invoked, about $0.0002 per session on Opus 5.5. 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 2026-10-03.
Other skills, from other repositories
speclite-canonical-source-governance-runner
A change-governance workflow for SpecLite's canonical source files and the documents, tests, hooks, and packaging files derived from them. It reads the change and records whether each affected area needs updating.
project-flow-ops
CIEL's coordination layer for triaging issues, PRs, and tasks across GitHub and internal tracking systems (e.g., Linear).
proposal-filing
File a new Lorekeeper proposal ticket — create markdown, create GitHub issue, commit, push. Use when requesting a new feature, filing a bug, or submitting a product idea.
gsd-quick-batch
Batch several /gsd-quick-shaped tasks together — planned, dispatched, and merged as one run.
gsd-pause-work
Create context handoff when pausing work mid-phase.
mcore-split-pr
Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.