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 skills/volcengine/ark-cli/arkcli-train-finetunenpx skills add volcengine/ark-cli --skill arkcli-train-finetunegit clone --depth 1 https://github.com/volcengine/ark-cliWhat 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 | $0.00114 | $0.02174 |
| Opus 5 | $0.00057 | $0.01087 |
| Sonnet 5 | $0.00023 | $0.00435 |
| Haiku 4.5 | $0.00011 | $0.00217 |
Grade A, and why
arkcli-train-finetune 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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ArkCLI 精调训练
先读取 ../arkcli-shared/SKILL.md，遵循认证、输出、安全和二次确认规则。
适用场景与能力边界
- 创建精调任务：读取
references/create.md - 列出或筛选任务：读取
references/list.md - 查询、观察或操作一个指定任务：读取
references/manage.md - 根据指标选择 step、导出产物并部署：读取
references/export-deploy.md - 不直接管理数据集，但可以使用本地文件、TOS URL、
ds-*/dsv-*引用和模型支持的 preset；需要创建或维护 Dataset 时转../arkcli-datasets/SKILL.md。 - 普通训练 Dataset 默认使用
--train-dataset（Multiplier=1）；需要重复引用、倍率或采样数时改用可重复的--train-path。每项最多设置multiplier或sample_count之一，均不设置时仍默认Multiplier=1。preset 必须在inject_multiplier与inject_sample_count中二选一。 - 训练产物的指标分析和 artifact export 由本 skill 编排；custom model 详情、可部署版本准备和 Endpoint 创建必须按模型仓库及部署 skill 执行。
- 不把 Raw API 或精调 SDK 当默认入口。
只加载当前任务需要的 reference。不要为了熟悉全部命令一次性读取所有文件。
反唤起信号
- 只管理数据集而不涉及精调任务 → 使用数据集能力，不要加载本 skill。
- 只查询公共基础模型目录 → 使用
../arkcli-models/SKILL.md。 - 只管理已有推理 Endpoint → 使用
../arkcli-infer-endpoint/SKILL.md。 - 纯登录或 profile/config 排障 → 分别使用
../arkcli-auth/SKILL.md或../arkcli-config/SKILL.md。 - 不要把 Raw API 或精调 SDK 当作默认入口；只有产品命令无法表达任务且用户确认 fallback 后才进入扩展流程。
指定任务的精确范围诊断
- 用户给出
mcj-*并询问任务状态、查不到原因、日志或 trajectory 时，必须加载本 skill 并读取references/manage.md。 - “这个任务怎么查不到”首先在当前 active profile / project / region 对原始 ID 执行
arkcli train finetune get <mcj-id>，再按该权威 API 的原始结果解释。 mcj-*是不透明资源 ID。不得根据日期片段、后缀单词或臆测的哈希格式断言 ID 无效，也不得改写用户给出的 ID。- 用户要求不切环境或只查指定任务时，禁止执行
train finetune list、扫描其他任务、切换 profile/project/region，或查询其他账号。目标get失败时保留错误 code、message 和 request ID；只有用户另行授权后才能扩大范围。 - 用户要求把指定 MCJ 的日志保存到本地路径时，第一条业务命令就是
arkcli train finetune logs <mcj-id> --output <path>。不得先 list 全部任务或用脚本遍历；目标命令失败时原样报告，不建议切环境。 - 用户要求指定 MCJ 的完整 rollout trajectory 时，直接执行
arkcli train finetune trajectory list <mcj-id> --full。不存在arkcli train trajectory路径；无轨迹或未开启记录时保留原错误，不探索 profile、MCP 或其他任务。 logs --follow仅在任务活跃且可能继续产生日志时持续轮询；任务已终态时输出当前快照后自动退出，轮询中发现终态且无新日志也会退出。不要再用外部 timeout 作为正常终止机制。pause与resume是明确的可逆关系：pause将运行任务置为Paused，resume用于恢复Paused；后端允许时也可用resume重试Failed/Terminated，以当前 API 结果为准。
What ships with it
9 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.
- 2d ago First seen · 106 lines · 114 tokens per session scan A 0ddf08929b51
arkcli-train-finetune is a skill published in the GitHub repository volcengine/ark-cli (112 stars, last pushed 5d ago), licensed Apache-2.0. It adds 114 tokens to every session and 2,174 once invoked, about $0.0006 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 skills, from other repositories
happiness-skill
当用户问「怎么才能更幸福/为什么得到了还不满足/怎么减少焦虑」时调用。 核心理念: 幸福是缺憾感清空的默认状态, 是可训练的技能; 欲望是与自己的契约(得到前不快乐), 同时只留一个重大欲望; 活在当下。 不适用于: 临床抑郁等需要专业治疗的场景(本书方法不能替代医疗)。 Triggers: 幸福/不快乐/欲望/焦虑/知足/活在当下/happiness/desire/anxiety.
create-business
Guided journey from raw idea to a validated, positioned, priced business with a chosen beachhead. Orchestrates ten skills phase by phase - jobs-to-be-done, mom-test, design-sprint, lean-startup, good-strategy-bad-strategy, blue-ocean-strategy, obviously-awesome, hundred-million-offers, monetizing-innovation…
jobs-to-be-done
Discover what customers truly need by analyzing the "job" they hire your product to do. Use when the user mentions "customer discovery", "why customers churn", "what job does this solve", "competing against luck", "product-market fit", "switching behavior", "milkshake moment", or "functional vs emotional jobs". Also…
asc-subscription-localization
Bulk-localize subscription, subscription-group, and in-app purchase display names across App Store locales using asc, including API 4.4.1 version-scoped v2 resources. Use when filling or updating subscription/IAP names and descriptions without App Store Connect UI work.
pcbway
PCBWay PCB fabrication and assembly — turnkey/consigned assembly, design rules, ordering workflow. Alternative to JLCPCB for manufacturing. Use with KiCad. Use this skill when the user mentions PCBWay, needs turnkey assembly (PCBWay sources parts by MPN), has parts not available on LCSC, needs assembled boards with…
scienceworld-growth-focuser
Use when you have planted a seed or need to track a plant's growth stage (sprouting, flowering, reproduction). Applies the 'focus on' action to a specific plant or biological entity to signal intent and monitor its development. Trigger after planting or when you need to observe life cycle progression in the…