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 Kihara777/NixKits --skill agent-self-maintenancegit clone --depth 1 https://github.com/Kihara777/NixKitsWrote 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/kihara777/nixkits/agent-self-maintenance)<a href="https://agentmods.dev/skills/kihara777/nixkits/agent-self-maintenance"><img src="https://agentmods.dev/badge/skills/kihara777/nixkits/agent-self-maintenance/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/kihara777/nixkits/agent-self-maintenance"><img src="https://agentmods.dev/badge/skills/kihara777/nixkits/agent-self-maintenance.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 | $0.00087 | $0.01432 |
| Opus 5.5 | $0.00035 | $0.00573 |
| Sonnet 5.5 | $0.00017 | $0.00286 |
| Haiku 4.5 | $0.00009 | $0.00143 |
Grade A, and why
agent-self-maintenance 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 3d 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 — 115 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent 自我维护：五个会失败的小机制
为什么需要它们
一个会长期工作的 agent 最容易发生的不是「做错」，而是**「没人问」**：
- 日记停在某一天，没人发现；
- 自检红了三天，没人知道；
- 事实库写着「上游已迁移」，而实际没有；
- 「她说的」和「我量的」被混成一句结论。
这五件事的共同点：不做也不会报错。所以每个机制都必须有一个「会失败」的形态， 并且要能回答「谁在跑它」。
五个机制
1. 日记闸门（journal gate）
判据（两条，都从本 session 自己的事件日志读）：
- 做过事：本 session 里出现过
git commit； - 没记事：
journal/<当日>.md不存在或未被本 session 修改。
两条同时成立 → 收尾时提醒一次。是门，不是自觉。
边界：只在「有 journal 惯例的仓库」启用；没配置就安静——不要每个 session 都念 「你没写日记」，那会变成噪音，而噪音会被忽略。
2. 开局对账（journal catchup）
闸门有一个结构性缺口：它只在 session 里活着。没有 session 的工作（终端里改的、 定时任务动的）它看不见。所以开局要读一次账：
| 输入 | 要回答的问题 |
|---|---|
| 自检收据 | 上次自检是什么时候？红了哪几项？ |
| 日记 | 上次日记是哪天？之后有没有提交？ |
| 事实库 | 有没有过期的事实？（每条事实带日期） |
三种状态不合并：ok / red / stale。「没跑过」必须说出来。
3. 事实核查（facts check）
事实库里的每一条都是活标本：带日期、带证据、带探针。
- 探针：一条能重新量出该事实的命令；
- 第三种状态：
无法核验（探针跑不了）——它既不是通过也不是失败； - 抄来的与量到的分开记：
source: upstream-docvssource: measured。
4. 自我观察（inquiry）
每一条观察回答三个问题：
- 这是她说的（引用 + 时间）？
- 还是我量的（命令 + 输出）？
- 还是不确定（明确标注）？
条目必须带日期与证据。不要在一句话里混用两种来源。
5. 自检收据（selfcheck receipt）
自检脚本每跑一次写一份收据（时刻、rev、失败项）。判据有五种状态：
| 状态 | 含义 |
|---|---|
ok |
刚跑过且全绿 |
red |
跑过但有失败项 |
stale |
太久没跑（阈值由仓库定） |
missing |
没有收据 |
unreadable |
收据在但读不出来 |
只有 ok 是安静的；其余四种都要在开局说出来。
边界：机制搬走，数据留在原地
| 东西 | 归属 |
|---|---|
| 五个机制的判据与形态 | 技能 / 预设（可移植） |
| 事实条目、观察条目、预算阈值、自检阈值 | 仓库自己的数据（不随机制搬） |
收据文件（state/selfcheck.json） |
仓库自己的产物（不进版本库） |
为什么这条重要：机制是「怎么判断」，数据是「判断什么」。把数据随机制一起下放， 等于替使用者断言了他们的事实——而那些事实第二天就会过期。
适配层：本仓的东西写在本仓
通用层只写「机制长什么样」，具体落点属于各仓库自己的适配层。分法是一句话：
| 写在这里（通用） | 写在适配层（各仓自己的） |
|---|---|
| 判据的形态（「做过事」+「没记事」两条同时成立才提醒） | 判据读哪个文件、用文件 mtime 还是 git commit |
| 五种状态的语义（ok / red / stale / missing / unreadable） | 阈值是多少、收据文件叫什么名字 |
| 「数据不随机制搬走」这条边界 | 事实库、观察库、预算各自放在哪 |
判据只有一句：换个仓库还成立吗？ 成立写这里，不成立写适配层。
写成适配层时至少要说清三件事：实现文件在哪、谁在跑它、哪些数字必须现取 （写死的数字第二天就会变成假话）。找不到「谁在跑」那一栏，就先别写——那说明它还是 说明书，不是机制。
最小落地
不要求任何特定 harness。三个可移植形态：
- 钩子（harness 支持时）：session 结束 / 开始时触发；
- 收尾命令：
make check或scripts/maintain.sh里串一遍； - 纯人工 + 收据：至少让「没跑」留下痕迹。
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.
- 3d ago First seen · 115 lines · 87 tokens per session scan A 4e21f1f8bbdd
agent-self-maintenance is a skill published in the GitHub repository Kihara777/NixKits (29 stars, last pushed yesterday), licensed MIT. It adds 87 tokens to every session and 1,432 once invoked, about $0.0003 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 2026-10-21.
Other skills, from other repositories
Cortex
Operate Cortex, the LifeOS memory system — the typed Knowledge Archive (People, Companies, Ideas, Research with typed related: links) plus recall of prior work sessions, ISAs, and conversations. Search, add, harvest, develop, ingest, distill, graph-navigate, recall. USE WHEN cortex, knowledge, knowledge base, search…
wiki-dashboard
Create dynamic, queryable Obsidian dashboard views using Bases or Dataview. Use for auto-updating tables, indexes, or structured views of vault content; not for ordinary question answering.
Migrate
Intakes external content, classifies chunks against LifeOS taxonomy, commits with provenance. Sources: .md/.txt, stdin, LifeOS dirs, CLAUDE.md/Cursor/OpenAI Custom Instructions, Obsidian/Notion/Apple Notes exports. MigrateScan classifies → routing table. MigrateApprove with…
relevance-coarse-filter
Cheap, high-recall first-pass filter that removes obvious junk from a detector candidate pool before expensive story-origin research and PR judgment. Decides keep, monitoronly, or reject — never ranks, writes angles, verifies dates, or decides whether to pitch.
session-compact
AI DevKit · Compact a historical AI coding session with Jev when context is long, work is handed off or resumed, or durable continuation facts and memory candidates need extraction.
company-brain
Your team's shared, AI-ready knowledge base — people, companies, meetings, SOPs, and decisions structured so an agent can answer on your team's behalf. Team-scope sibling to second-brain — personal notes go there, not here. Modes — capture, compile (wiki pages + INDEX.md), query (trust-weighted, saved to outputs/)…