Turn expiring AI tokens into durable personal output — ideas, research, code, tests, reviews, refactors, and data pipelines.
将即将过期的 AI Token 转化为可沉淀的个人产出——创意、研究、代码、测试、审查、重构与数据管道。
- What is it? | 这是什么？
- The Burn Pipeline | 燃烧流水线
- Family Members | 家族成员
- Quick Start | 快速开始
- Detailed Usage | 详细使用方法
- Recommended Workflow | 推荐工作流
- Folder Structure | 文件夹结构
- Image Assets Guide | 插图素材指南
burn-my-tokens is a family of AI agent skills designed for Claude Code. It solves a common problem: your AI coding plan quota is about to reset, and you have thousands of tokens left that will simply vanish.
Instead of letting them expire, burn-my-tokens autonomously converts those remaining tokens into tangible, durable outputs:
- 🧠 Structured ideas with market validation
- 📊 Deep research reports with citations
- 🚀 Runnable MVP projects with PRDs and tests
- 🧪 Test coverage for existing codebases
- 🔍 Code quality reviews with health scores
- 🛠️ Safe refactors with diff logs
- 📈 Clean datasets with visualizations
All skills support unattended self-iterative burning with graceful stop, auto-resume via CronCreate, and pure text interaction — no complex setup required.
burn-my-tokens 是一套为 Claude Code 设计的 AI Agent Skill 家族。它解决了一个常见问题：你的 AI Coding Plan 额度即将重置，还有大量 Token 即将白白浪费。
burn-my-tokens 能够自主地将剩余 Token 转化为可沉淀的实际产出：
- 🧠 结构化创意——附带市场验证
- 📊 深度研究报告——附带引用来源
- 🚀 可运行的 MVP 项目——附带 PRD 和测试
- 🧪 现有代码库的测试覆盖
- 🔍 代码质量审查——附带健康评分
- 🛠️ 安全重构——附带 diff 日志
- 📈 清洗后的数据集——附带可视化
所有 Skill 均支持无人值守的自主迭代燃烧、通过 CronCreate 实现的优雅停止与自动恢复，以及纯文本交互——无需复杂配置。
┌─────────────────────────────────────────────────────────────────────────────┐
│ burn-my-tokens-X (Full Pipeline) │
│ 完整流水线（7 阶段） │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ [1] Idea Generation [2] Deep Research │
│ 创意生成 深度研究 │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────┐ ┌─────────────┐ │
│ │ domain │ │ research │ │
│ │ analysis │ │ outline │ │
│ │ raw ideas │ │ findings │ │
│ │ idea cards │ │ synthesis │ │
│ │ top picks │ │ report │ │
│ └─────────────┘ └─────────────┘ │
│ │ │ │
│ ▼ ▼ │
│ [3] MVP Generation [4] Data Engineering │
│ MVP 生成 数据工程 │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────┐ ┌─────────────┐ │
│ │ PRD │ │ data profile│ │
│ │ src/ │ │ cleaning │ │
│ │ tests/ │ │ pipeline │ │
│ │ verification│ │ plots │ │
│ └─────────────┘ └─────────────┘ │
│ │ │ │
│ ▼ ▼ │
│ [5] Test Generation [6] Code Review │
│ 测试生成 代码审查 │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────┐ ┌─────────────┐ │
│ │ coverage gap│ │ analysis │ │
│ │ test strategy│ │ review plan │ │
│ │ test files │ │ per-module │ │
│ │ validation │ │ master review│ │
│ └─────────────┘ └─────────────┘ │
│ │ │ │
│ └────────────┬─────────────┘ │
│ ▼ │
│ [7] Refactoring │
│ 代码重构 │
│ │ │
│ ▼ │
│ ┌─────────────┐ │
│ │ smell report│ │
│ │ refactor plan│ │
│ │ diff logs │ │
│ │ validation │ │
│ └─────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────┐ │
│ │ LOOP BACK ─────►│ Iteration 2+ │
│ │ 循环迭代 │ 第 2+ 轮迭代 │
│ └─────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Illustration:
assets/pipeline_diagram.png
| Skill | Purpose | Output | Code? |
|---|---|---|---|
burn-my-tokens-idea |
Structured ideation & concept evaluation | Domain analysis, idea cards, top picks, hybrid concepts | ❌ No |
burn-my-tokens-research |
Deep domain research with citations | Research outline, findings, synthesis, master report | ❌ No |
burn-my-tokens-MVP |
End-to-end MVP project generation | PRD, src/, tests/, requirements.txt, run verification |
✅ Yes |
burn-my-tokens-data |
Data engineering & visualization | Cleaned datasets, pipeline scripts, plots, profiles | ✅ Yes |
burn-my-tokens-testgen |
Automated test coverage generation | pytest files, coverage delta report | ✅ Yes |
burn-my-tokens-review |
Structured code review & health scoring | Analysis report, per-module reviews, master review | ❌ No (read-only) |
burn-my-tokens-refactor |
Safe code refactoring with validation | Smell report, refactor plan, diff logs | ✅ Yes (modifies code) |
burn-my-tokens-X |
Full pipeline orchestrator — runs all 7 stages end-to-end with budget allocation and iterative refinement | Complete iteration reports, stage summaries, pipeline report | ✅ Yes |
| Skill | 用途 | 产出 | 是否生成代码 |
|---|---|---|---|
burn-my-tokens-idea |
结构化创意与概念评估 | 领域分析、创意卡片、精选方案、混合概念 | ❌ 否 |
burn-my-tokens-research |
带引用的深度领域研究 | 研究大纲、发现、综合、主报告 | ❌ 否 |
burn-my-tokens-MVP |
端到端 MVP 项目生成 | PRD、src/、tests/、requirements.txt、运行验证 |
✅ 是 |
burn-my-tokens-data |
数据工程与可视化 | 清洗后的数据集、管道脚本、图表、画像 | ✅ 是 |
burn-my-tokens-testgen |
自动化测试覆盖生成 | pytest 文件、覆盖率变化报告 | ✅ 是 |
burn-my-tokens-review |
结构化代码审查与健康评分 | 分析报告、模块级审查、主审查报告 | ❌ 否（只读） |
burn-my-tokens-refactor |
带验证的安全代码重构 | 异味报告、重构计划、diff 日志 | ✅ 是（修改代码） |
burn-my-tokens-X |
完整流水线编排器——端到端运行全部 7 个阶段，支持预算分配与迭代优化 | 完整迭代报告、阶段摘要、流水线报告 | ✅ 是 |
- Claude Code installed and authenticated
- A Claude Code project with
.claude/skills/or.claude/rules/directory - Python 3.10+ (for code-generating skills)
pytest,pytest-cov(optional, for test generation)
-
Copy the desired skill directory into your Claude Code project:
cp -r burn-my-tokens-MVP /path/to/your/project/.claude/skills/
-
Or place the
skill.mdcontent directly into your Claude Code custom skills. -
Start burning:
/burn-my-tokens-MVP
All skills share a unified tier system:
| Tier | Approximate Tokens | Best For |
|---|---|---|
10k |
~10,000 | Quick scan / lightweight output |
100k |
~100,000 | Standard depth (1 module/topic) |
1M |
~1,000,000 | Deep burn (3-4 modules/topics) |
10M |
~10,000,000 | Massive burn (8-12 modules/topics) |
burn |
Unlimited | Burn until stopped or quota exhausted |
Note: All tiers include graceful stop (stop command) and auto-resume via CronCreate.
Trigger: /burn-my-tokens-idea
Inputs:
- Target domain (e.g., "AI tools for veterinary clinics")
- Constraints (budget, tech, time, regulatory)
- Idea type (product / feature / business / content / all)
- Target audience (optional)
- Emphasis preferences (competitive differentiation, tech novelty, etc.)
Workflow:
- Domain analysis via WebSearch (market landscape, pain points, underserved segments)
- Parallel angle ideation (max 2 subagents)
- Idea evaluation with 5-axis scoring (feasibility, novelty, market size, differentiation, personal fit)
- Top pick refinement into structured concept briefs
- Cross-pollination (hybrid idea generation)
Outputs:
burn-my-tokens-idea_output/<task_name>/
├── domain_analysis.md
├── raw_ideas/
│ └── <angle>.md
├── IDEA_CARDS.md
├── TOP_PICKS.md
└── IDEA_BURN_REPORT.md
Burn Contract: Never generates source code. Output is exclusively markdown documents.
Trigger: /burn-my-tokens-research
Inputs:
- Target domain or field
- Specific research questions (optional)
- Output depth (brief / standard / comprehensive)
- Analysis sections (market / technology / competitive / trends / all)
Workflow:
- Research scoping (core questions, subtopics, priority mapping)
- Parallel information gathering (max 2 subagents per subtopic)
- Cross-reference & validation (source reliability tiers)
- Synthesis & pattern identification
- Master report generation with inline citations
- Validation audit (coverage, citation completeness, source diversity)
Outputs:
burn-my-tokens-research_output/<task_name>/
├── research_outline.md
├── findings/
│ └── <subtopic>.md
├── synthesis_notes.md
├── RESEARCH_REPORT.md
├── validation_report.md
└── RESEARCH_BURN_REPORT.md
Source Reliability Tiers:
- Tier 1: Primary sources (SEC filings, academic papers, government stats)
- Tier 2: Reputable media (Reuters, Bloomberg, analyst reports)
- Tier 3: Blogs, forums, opinions (flagged, not used for quantitative claims)
Trigger: /burn-my-tokens-MVP
Inputs:
- Tier selection
- Existing project directories to analyze (optional)
- Technical domain or desired direction (optional)
- Execution mode (semi-auto / full-auto)
Workflow:
- Analyze existing projects (read PRD.md, PROGRESS.md, README.md)
- Generate direction list (2-3 niche directions via WebSearch)
- Self-critique (eliminate overlaps, infeasible concepts)
- Critique-agent review (scoring 1-10, priority ranking)
- Present & confirm (or auto-select in full-auto mode)
- Launch subagents (max 2 parallel) for L3 MVP generation
- Each subagent follows: Research → PRD → Core Dev → Unit Tests → Verification → PROGRESS.md
Subagent L3 MVP Steps:
| Step | Budget | Output |
|---|---|---|
| Web Research | 15-20% | research.md |
| Write PRD | 15-20% | PRD.md |
| Core Development | 40-50% | src/ directory |
| Unit Tests | 10-15% | tests/ directory |
| Verification Run | 5-10% | Run log summary |
| PROGRESS.md | 5% | PROGRESS.md |
Outputs:
burn-my-tokens-MVP_output/<project_name>/
├── research.md
├── PRD.md
├── src/
├── tests/
├── requirements.txt
├── main.py
├── PROGRESS.md
└── .burn_state.json
Trigger: /burn-my-tokens-data
Inputs:
- Data source (file path / API endpoint / "find data online")
- Goal (clean / transform / visualize / collect / analyze / all)
- Output format (CSV / JSON / SQLite / HTML / plots / all)
Workflow:
- Data discovery / ingestion (auto-detect format, profile)
- Cleaning strategy (impact-scored issue prioritization)
- Execute pipeline (parallel subagents, max 2)
- Validation (quality metrics, regression detection)
Outputs:
burn-my-tokens-data_output/<task_name>/
├── data/
│ └── raw/
├── outputs/
│ ├── cleaned_<dataset>.csv
│ └── plots/
├── src/
│ └── pipeline_<dataset>.py
├── data_profile.md
├── cleaning_plan.md
└── DATA_BURN_REPORT.md
Trigger: /burn-my-tokens-testgen
Inputs:
- Codebase directory
- Target coverage percentage (default: 80%)
Workflow:
- Coverage analysis (
pytest --covor static fallback) - Test strategy (risk-tier prioritization, test type selection)
- Generate tests (parallel subagents, max 2)
- Run & fix (degrade on failure: property-based → example-based → simple happy path)
- Validation (full suite run, coverage delta tracking)
Test Types Generated:
- Unit tests with type hints and Google-style docstrings
- Boundary tests (empty, None, max, negative)
- Error path tests (expected exceptions, invalid inputs)
- Property-based tests (Hypothesis, optional)
- Integration tests with mocked dependencies
Outputs:
burn-my-tokens-testgen_output/
├── coverage_gap.md
├── test_strategy.md
├── tests/
│ └── test_<module>.py
└── TEST_BURN_REPORT.md
Trigger: /burn-my-tokens-review
Inputs:
- Codebase directory
- Focus areas (performance / readability / architecture / security / all)
Workflow:
- Code analysis (complexity hotspots, coupling, documentation gaps, type hint gaps, potential bugs, performance issues, security issues)
- Issue prioritization (severity matrix: likelihood × impact)
- Generate reviews (parallel subagents, max 2)
- Consolidation (master review, code health score, top 10 issues)
Severity Matrix:
| Likelihood \ Impact | High | Medium | Low |
|---|---|---|---|
| High | Critical | High | Medium |
| Medium | High | Medium | Low |
| Low | Medium | Low | Low |
Module Health Score:
module_health_score = 100 - (
critical_issues × 10 +
high_issues × 5 +
medium_issues × 2 +
low_issues × 0.5 +
cc_penalty +
docstring_penalty +
type_hint_penalty
)
Outputs:
burn-my-tokens-review_output/
├── analysis_report.md
├── review_plan.md
├── review_reports/
│ └── <module>_review.md
├── MASTER_REVIEW.md
└── REVIEW_BURN_REPORT.md
Safety: Read-only. Never modifies source code.
Trigger: /burn-my-tokens-refactor
Inputs:
- Codebase directory
- Focus area (complexity / duplication / naming / all)
Workflow:
- Smell detection (radon cc/mi, jscpd, pylint, ruff; AST-based fallback)
- Prioritization (impact × ease_of_fix)
- Baseline test run (record before any changes)
- Execute refactoring (parallel subagents, max 2)
- Validation loop (re-run tests, re-detect smells, compare metrics)
Supported Refactorings:
- Extract functions / methods
- Rename variables / functions / classes
- Remove duplicated code (extract to shared utility)
- Simplify nested conditionals (guard clauses, early returns)
- Reorganize imports
- Remove dead code
Failure Handling (Conservative Retry):
- First failure: Rollback, extract smaller functions only
- Second failure: Rollback, rename and import cleanup only
- Third failure: Rollback, skip module, log reason
Outputs:
burn-my-tokens_output/refactor_<project_name>/
├── smell_report.md
├── refactor_plan.md
├── refactor_logs/
│ └── <module>.diff
└── REFACTOR_BURN_REPORT.md
Trigger: /burn-my-tokens-X
Inputs:
- Tier selection
- Project domain or idea area
- Total pipeline budget (optional)
- Stages to skip (optional)
Budget Allocation (Default):
| Stage | Ratio | Purpose |
|---|---|---|
| Idea | 15% | Domain analysis, ideation, evaluation |
| Research | 10% | Competitive landscape, citations |
| MVP | 30% | PRD, core dev, tests, verification |
| Data | 10% | Dataset discovery, cleaning, profiling |
| TestGen | 10% | Coverage analysis, test generation |
| Review | 10% | Code analysis, structured reviews |
| Refactor | 15% | Smell detection, safe refactoring |
Workflow:
- Initialize state & output structure
- CronCreate resume task
- Execute Stage 1 → Stage 2 → ... → Stage 7
- After Stage 7: write iteration summary, check loop condition
- If budget remains: start Iteration 2+ with carried-over context
Iteration Refinement Strategy:
- Iteration 1: Greenfield — generate new idea, build from scratch
- Iteration 2+: Refinement — deepen research, extend MVP, enrich datasets, increase coverage, fix issues, reduce smells
Outputs:
burn-my-tokens-X_output/<pipeline_name>/
├── .burn_state.json
├── PIPELINE_REPORT.md
├── iteration_1/
│ ├── stage_1_idea/
│ ├── stage_2_research/
│ ├── stage_3_mvp/
│ ├── stage_4_data/
│ ├── stage_5_testgen/
│ ├── stage_6_review/
│ └── stage_7_refactor/
└── stage_summaries/
└── iteration_1_summary.md
/burn-my-tokens-idea → Explore niche opportunities
探索细分机会
↓
Select top idea from TOP_PICKS.md
从 TOP_PICKS.md 中挑选最佳创意
↓
/burn-my-tokens-X → Run full pipeline
运行完整流水线
↓
Review PIPELINE_REPORT.md
审查 PIPELINE_REPORT.md
/burn-my-tokens-review → Identify all issues
识别所有问题
↓
/burn-my-tokens-testgen → Add missing tests
补充缺失测试
↓
/burn-my-tokens-refactor → Apply safe fixes
应用安全修复
↓
/burn-my-tokens-review → Verify improvements
验证改进效果
/burn-my-tokens-research → Deep domain research
深度领域研究
↓
/burn-my-tokens-data → Collect & analyze datasets
收集与分析数据集
↓
/burn-my-tokens-MVP → Build MVP based on insights
基于洞察构建 MVP
github_submit/
├── README.md
├── burn-my-tokens-MVP/
│ ├── skill.md # End-to-end MVP generation
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-X/
│ ├── skill.md # Full pipeline orchestrator (7 stages)
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-data/
│ ├── skill.md # Data engineering & visualization
│ ├── skill_dataset_discovery.md # Dataset discovery variant
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-idea/
│ ├── skill.md # Structured ideation & concept evaluation
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-refactor/
│ ├── skill.md # Safe code refactoring
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-research/
│ ├── skill.md # Deep research with citations
│ └── PRD.md # Product Requirements Document
├── burn-my-tokens-review/
│ ├── skill.md # Structured code review
│ └── PRD.md # Product Requirements Document
└── burn-my-tokens-testgen/
├── skill.md # Automated test generation
└── PRD.md # Product Requirements Document
All tiers multiplied by 10x to match increased token budgets:
| Old Tier | New Tier | Approximate Tokens |
|---|---|---|
1k |
10k |
~10,000 |
10k |
100k |
~100,000 |
100k |
1M |
~1,000,000 |
1M |
10M |
~10,000,000 |
Default tier changes:
- Default tier for all individual skills (except X):
1M - Default tier for
burn-my-tokens-Xorchestrator:10M
Allocation changes:
- All stage budgets and pipeline allocations multiplied by 10x accordingly
- Per-stage hard cap in X:
100k→1M
MIT License — feel free to use, modify, and distribute.
MIT 许可证——可自由使用、修改和分发。
This is an open skill family for the Claude Code ecosystem. Contributions welcome:
- New burn skills (e.g.,
burn-my-tokens-docs,burn-my-tokens-design) - Language adaptations
- Pipeline optimizations
- Bug fixes and edge case handling
这是面向 Claude Code 生态的开放 Skill 家族。欢迎贡献：
- 新增燃烧 Skill（如
burn-my-tokens-docs、burn-my-tokens-design） - 多语言适配
- 流水线优化
- Bug 修复与边界情况处理