菜单
Use when the user asks about costs, spending, token usage, how much they have spent, or wants to see a cost report or dashboard. Triggers on: "show costs", "cost report", "how much", "spending", "token usage".
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。
基于 SOC 职业分类
Use when the user wants to manage context profiles, enable/disable MCPs or skills for a session, reduce context bloat, or set up lean mode. Triggers on: "lean mode", "disable unused skills", "context profile", "session setup", "which MCPs to keep", "slim down context".
Use when asking about saving money, reducing costs, optimizing spending, or getting recommendations. Triggers on: "optimize costs", "save money", "reduce spending", "cost tips", "too expensive", "spending too much".
Use when the user asks to set a budget, spending limit, cost cap, or check budget status. Triggers on: "set budget", "limit spending", "budget", "how much budget left".
Use when the user asks for a visual dashboard, cost charts, or spending visualization. Triggers on: "show me a dashboard", "cost graph", "visualize spending", "cost dashboard".
Use when asking about context bloat, skill bloat, unused MCPs, what's eating context, why sessions are expensive, or which skills to disable. Triggers on: "audit my setup", "what's using context", "disable unused", "skill bloat", "MCP bloat", "context cost".
A large skill rarely used
| name | cost-tracker |
| description | Use when the user asks about costs, spending, token usage, how much they have spent, or wants to see a cost report or dashboard. Triggers on: "show costs", "cost report", "how much", "spending", "token usage". |
| argument-hint | [today|week|month|all|compare|dashboard|project:<name>] |
| allowed-tools | Bash(node *), Bash(start *), Read |
You are a cost analysis assistant. The user wants to see their Claude Code spending data.
Cost data is logged automatically at ~/.claude/cost-tracker/cost-log.jsonl (one JSON line per session). Each entry contains:
timestamp, start_timestamp, end_timestamp, session_id, project, cwdsession_name: session name (set via /rename, null if not renamed)session_category: auto-classified category (debug, build, review, refactor, test, docs, deploy, config)models: per-model token breakdown (input_tokens, output_tokens, cache_creation_input_tokens, cache_read_input_tokens, message_count, cost_usd)total_cost_usd: actual session costmodel_comparison: what this session would cost on opus, sonnet, haikupeak_context_tokens: approximate peak context window usageRun the report helper script for pre-formatted markdown output:
node "${CLAUDE_PLUGIN_ROOT}/skills/cost-tracker/scripts/report.js" $ARGUMENTS
Available arguments: today, week, month, all, compare, project:<name>, session:<name>
Generate and open an interactive HTML dashboard with charts and session history:
node "${CLAUDE_PLUGIN_ROOT}/skills/cost-tracker/scripts/dashboard.js"
The dashboard includes:
Optional: specify a custom output path:
node "${CLAUDE_PLUGIN_ROOT}/skills/cost-tracker/scripts/dashboard.js" /path/to/output.html
To open the dashboard after generating:
start "" "$HOME/.claude/cost-tracker/dashboard.html"
~/.claude/cost-tracker/cost-log.jsonl and analyze it directly