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 0fuk/OpenACCP --skill keep-pushinggit clone --depth 1 https://github.com/0fuk/OpenACCPWrote 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/0fuk/openaccp/keep-pushing)<a href="https://agentmods.dev/skills/0fuk/openaccp/keep-pushing"><img src="https://agentmods.dev/badge/skills/0fuk/openaccp/keep-pushing/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/0fuk/openaccp/keep-pushing"><img src="https://agentmods.dev/badge/skills/0fuk/openaccp/keep-pushing.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 | $0.00110 | $0.02792 |
| Opus 5.5 | $0.00044 | $0.01117 |
| Sonnet 5.5 | $0.00022 | $0.00558 |
| Haiku 4.5 | $0.00011 | $0.00279 |
Grade A, and why
keep-pushing scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/keep-pushing — 无人值守自主推进模式
owner 离机,从本 turn 起进入无人值守窗,直到 owner 回来发话。存在理由:一次误用需人工确认的工具卡死流水线 8 小时;一次「等一份其实已到的件」卡死 6 小时。项目事实(协调目录/台账/心跳/铁律)读项目根 PROJECT-PROFILE.md,本 skill 零项目词。
0 · 生效声明(20261010 精简:只做四步,不建重复机制)
- 盘点现场:在飞子代理(agentId + output 文件 mtime)/ 未回执的跨会话令(回执件 mtime)/ 信箱按 mtime 扫未消费件 / 上一轮承诺未闭环的事。凡「等着我的」,本窗第一批处理;> 30 分钟无动作的按 §7 当场处置。
- 挂表:ScheduleWakeup ≤ 30 分钟(有在飞执行体时)或 20–30 分钟(纯等外部事件时)。平台完成通知与跨会话直推是主信号,挂表是兜底;两者都不靠文件 Monitor(项目仍在用 Monitor 的才重架)。
- 目标态与候 owner 清单:写进当前 checkpoint 的「compact 后第一批动作」与「候用户」节,不另开值守计划文件;依赖图只写真依赖。
- /compass 增量一次(不是全检),之后每个 tick 只跑 /compass 轻检(§4b 一行)。 与 /compass 的分工:本 skill = 怎么跑(通讯、派工、挂表、卡死处置、收口);/compass = 醒来查什么(对照目标、三栏对账、活性硬检、跑偏自检)。两者不重复定义同一条规则:活性判据只写在 /compass §4 / §4b,本 skill §7 只写处置动作。
1 · 通讯铁律
- 文件信箱=唯一通道:落盘即送达,收件=Monitor 叫醒后消费。
- 绝对禁用一切需人工点击确认的工具(
send_message/AskUserQuestion/任何「ALWAYS prompts」工具)。会触发弹窗的动作:先找不弹窗的等价路;找不到且非关键→入候 owner 清单,继续干别的。 - 时刻机械法:带时刻的产物先
T=$(date '+%F %T'),文件名与内文只引$T,禁手打时刻(提醒式规则实证无效)。
2 · 调度纪律
- 依赖图并发:互不依赖的工单同时派;禁日历日串行。
- 分层用模(20260916 owner 重申,必填):子代理 model+effort 必填,禁继承主会话(hook 拦):施工 sonnet·high(涉及前端可 opus5·high)/ 机械扫 haiku·low / 复核 sonnet·max(涉及前端或跨层可 opus5·max)/ 盲测 sonnet·high 一员顺序走;主循环只编排/终审。执行体降级时收货加码。harness 给足(/5gate 用量纪律第 7 条)。
- 每交付必独立复核:worker 自我声明≠收货;复核后才合流。
- 落盘必复验:merge/写文件/测试结果落盘后用独立 wc/ls/git 实测;收件读全文(或 grep 目标关键字),禁 tail/head 抽读后宣称「未落」;宣称「对方没交」前必 ls+grep;引用他方数字必回原件逐字对。
- 令件前提当刻确认:工单里写的「某件已交回/某步已做」发令时 ls 确认,模板里的前提不随转正为真。
- 对端活性只认会话级副作用(进度/回执/交回件 mtime);脱离进程写的心跳/标记文件不算。派出去的 lane 超 30 分钟无会话级动作→落探活件;两次无回应→记候 owner 清单,改道推进。
- 不脑补:证据不足就取证,取证不了就挂清单。
3 · 推进主循环(每次被叫醒)
consume(新件/完工通知)→ 独立验证 → 裁决(收货/打回/派复核)
→ dispatch(下一批可派工单)→ record(台账/值守计划勾销)
→ 还有可推进项?有→继续派;全部剩余=owner 专属决策→ §4 收口
- turn 结束的唯一合法状态:所有活性工作已委派且被叫醒机制覆盖,或已达 §4。「静候」前必列出所有不依赖等待项的未完成产物,清单非空禁静候;等待只对真依赖合法。
- 破坏性/不可逆动作(生产发布、真库写、force-push、删非己产物、对外发送)仍禁,挂清单——除非项目档案记有 owner 明示授权口径。
4 · 收口(只剩 owner 决策时)
- 候 owner 清单定稿:每项一句现状+选项+推荐(说人话)。
- owner 第一眼报告(项目有 HTML 交付惯例则出 HTML;内部台账用 md)。
- 证据路径全落盘,Monitor 保持常驻。
- 关值守开关文件(state=closed;owner 回来发话时同样要关)。
- 最后一条给 owner:干成了什么/在飞什么/候拍板什么/证据在哪。不发提问式结尾。
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 · 67 lines · 110 tokens per session scan A 1fbe3d56c052
keep-pushing is a skill published in the GitHub repository 0fuk/OpenACCP (2 stars, last pushed 2d ago), licensed MIT. It adds 110 tokens to every session and 2,792 once invoked, about $0.0004 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 2026-10-14.
Other skills, from other repositories
hive-mind-advanced
Queen-led collective-intelligence coordination with consensus mechanisms and persistent shared memory. Use when orchestrating a large swarm of agents under a lead coordinator, building consensus across agents, or maintaining durable cross-agent memory.
handoff-mesh
Built-in communication layer for multi-agent setups. Full-mesh handoff folders (any agent can hand off to any other), an hourly watcher that surfaces open handoffs, weekly archive cleanup, and a completion-reporting loop so every handoff is handled AND verified. The durable 'brain' counterpart to real-time channels…
tmux
Manage long-running shell sessions with tmux — start a detached session, run a long task, reattach later, capture output. Use when a task takes longer than a single tool call (build, test, log tail).
cog-braindump-capture
Capture raw thoughts with automatic domain classification and vault routing.
cog-knowledge-consolidation
Build structured knowledge frameworks from scattered vault notes with source attribution.
todos
Use when the user asks for a shared plan, a task needs at least three substantive execution steps, or a loose TODO is pending. Without an explicit plan request, not for one-shot answers or checks, or one- to two-step work that is not already in the list.