菜单
Use when someone asks what their Claude Code work has cost — "what did that cost", "how much have we spent", "where did the tokens go", "which model is burning the budget", "is the cache actually helping" — and wants it attributed to the work rather than a session total. Routes to /audit:usage.
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/AleksandarBisevac/claude-plugins --skill audit-spend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。
Edit, split or share the CSS and JavaScript under plugins/audit/scripts/ui/ — files Python concatenates into one inline <style> and one inline <script> carrying code, in a self-contained page opened over file://. Covers the assembly contract and the byte-level pins in plugins/audit/tests/ that guard it, why the split must be an order-preserving cut rather than a regrouping, the index-slice assertions that make section-marker comments load-bearing, the counts that currently enforce duplication, choosing a feature a report may use (Baseline plus a file:// gate), one dialect and one shared layer across both surfaces, CSS token and naming conventions, and the browser gates that must stay green. Use when touching anything under scripts/ui/ or _ui_theme.py, when extracting a shared partial or module, when a rule or helper exists twice, or when a selftest pin goes red after a UI edit.
How Python is written in this repo — free functions over classes, structured returns instead of mutated arguments, comprehensions, no module state, and the stdlib-only, annotation-free, Python 3.8 dialect the AST lints enforce. Covers modular structure and the import-layer rule, DRY without copying a helper into a second file, what makes a function testable from a `--selftest`, readability rules (guard clauses, named predicates, why-not-what docstrings), and the anti-patterns that have actually bitten here. Use when planning, writing, reading, reviewing or refactoring any `.py` under plugins/audit/, when adding a helper or a new script, when deciding where a function belongs, or when the same logic is about to exist in two places.
Write Python whose failures are visible and checks that actually fire — distinct sentinels for success and error, filters that narrow to nothing without reading as "all clear", output never discarded on a non-zero exit, and selftest cases proven red before they are trusted (mutate the fix, mutate it the other way too, count occurrences rather than asserting presence, pick fixture values that tell the two versions apart). Use when adding or reviewing a `--selftest` case, when a guard or lint is added, when a function reduces observations to a number a threshold reads, when a bug is fixed and a regression case is written, or when a check has only ever been seen passing.
基于 SOC 职业分类
正在显示 SKILL.md
| name | audit-spend |
| description | Use when someone asks what their Claude Code work has cost — "what did that cost", "how much have we spent", "where did the tokens go", "which model is burning the budget", "is the cache actually helping" — and wants it attributed to the work rather than a session total. Routes to /audit:usage. |
Read ${CLAUDE_PLUGIN_ROOT}/commands/usage.md and follow it. This file only exists so the
question finds the command.
/audit:usage --backfill needs no manifest, no setup and no agents. It reads the
transcripts already on disk and attributes past spend by phase, task, model, author and
day. If someone is asking what things cost, run this — do not first propose installing a
pipeline. It is the cheapest true answer available and it uses data they already have.
/cost does not/cost and /stats are session-scoped totals. This is spend attributed to plan units —
which phase, which task, which model, which person, with cache economics and cost-per-task.
Native OpenTelemetry has no phase or task in its attribute set, so this is not a nicer
rendering of the same numbers; it is a different join.
audit-usage.py renders its own ASCII output. Print it verbatim. Laying the tables out
again costs tokens to answer a question about token cost, which is self-defeating — the
command file says this too, and it is the rule the rest of this plugin's output follows.
If you are asked something the ledger cannot support — whether a cheaper model would do, what the next phase will cost — check what the output actually says. The routing advisory stays silent without enough in-repo evidence, and projections are suppressed below a sample gate. That silence is the answer, not a gap to fill in with a price list.