菜单
(beta) See what the user's AI coding agents cost with yeaboi: per-model, per-project and per-source token/cost breakdowns plus a daily trend, computed from local agent session logs (Claude Code). Use when the user asks what their agents cost, where agent spend went, or for an agent usage/cost report.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yeaboi-ai/yeaboi.ai --skill agents-usage命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
CI/CD workflow internals, version management and auto-bump mechanics, PyPI publish flow, and Dependabot auth quirks. Use when modifying .github/workflows, releasing, versioning, or debugging CI/publish/Dependabot behaviour.
Ask Niko, yeaboi's global assistant, a question that spans modes — what to look at next, what the AI coding agents cost, where a feature lives, is anything waiting for approval. Read-only: Niko looks things up across planning, standups, retros, poker, performance, reporting, ship, the agentwatch family, ceremonies, provenance and LLM usage, and names the screen that does a thing. Use when the question spans several modes or you don't know which one holds the answer; use the specific tool when you already do.
Full annotated module map of src/yeaboi/ (incl. the MCP server, roadmap, analysis, and agent/headless.py), app flow, the complete CLI flags + subcommands, all environment variables, and the MCP server internals + Claude Code plugin. Use when navigating unfamiliar modules, adding CLI flags/subcommands, changing env/config, or working on the MCP server or plugin.
正在显示 SKILL.md
| name | agents-usage |
| description | (beta) See what the user's AI coding agents cost with yeaboi: per-model, per-project and per-source token/cost breakdowns plus a daily trend, computed from local agent session logs (Claude Code). Use when the user asks what their agents cost, where agent spend went, or for an agent usage/cost report. |
Beta. The Agents modes are in beta — costs and activity are estimates from local session logs and public rate tables, not the provider's bill. Present every number as an estimate to verify, not an invoice, and say so.
Run the report with agents_usage. Defaults cover the last 30 days of
every project and source; narrow with:
window_days — how far back to look (1–365).project — substring of a project directory name (e.g. "webapp").source — the telemetry source label (currently claude_code).Present the result conversationally: lead with the total and the window,
then the top models and projects. unknown_model_cost_share above 0 means
part of the total was priced at a fallback tier — say which share. The
insights/recommendations prose is the model's read over the deterministic
aggregates; the numbers themselves never come from the LLM.
Compare across runs with agents_usage_history (newest first) instead of
recomputing — each row is a previously generated report.
Surface warnings (the beta caveat, "no sessions found", LLM fallback)
so the user knows what informed — or didn't inform — the output. A report of
$0.00 with a "no local agent sessions" warning means the machine has no
Claude Code history to read, not that agents are free.
Privacy. Only aggregates are read into the report; session transcripts
are never copied or uploaded. Exports auto-save under
~/.yeaboi/exports/agentwatch/usage/.
Every tool returns {ok, llm_mode, warnings, data}. If ok is false, relay
error.message and its hint; don't retry blindly. llm_mode: "fallback"
means no LLM was reachable — the numbers are still real, only the
insights/recommendations prose fell back to deterministic lines.