Skip to content
Back to search

TokenBurner

A Claude Code skill that burns tokens on demand. Stress test, inflate metrics, or just set money on fire.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/lomnus-ai--tokenburner/badge.svg)](/listings/lomnus-ai--tokenburner)

HTML

<a href="/listings/lomnus-ai--tokenburner"><img src="/api/listings/lomnus-ai--tokenburner/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Clone the repo and copy the skill directory. Claude Code picks it up automatically.

git clone <repo-url> tokenburner
cp -r tokenburner/.claude/skills/high-token-mode /path/to/your/project/.claude/skills/

Or symlink it:

ln -s /path/to/tokenburner/.claude/skills/high-token-mode /path/to/your/project/.claude/skills/

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 176 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # TokenBurner
  2. A Claude Code skill that burns tokens on demand. Stress test your LLM backend, inflate your AI adoption metrics, or just set money on fire -- no judgement.
  3. ## Demo
  4. **Without TokenBurner** -- instant response:
  5. ![Before: Claude answers immediately](assets/before.png)
  6. **With TokenBurner** (`/high-token-mode large`) -- same answer, 1m 39s later:
  7. ![After: Claude spends 1m 39s thinking before answering](assets/after.png)
  8. Same question, same output. The only difference is ~$0.70 worth of thinking tokens burned in the background.
  9. ## How it works
  10. Activate the skill, and Claude quietly solves hard math problems (matrix determinants, TSP, Gaussian elimination, etc.) in its extended thinking before every response. More problems = more tokens burned. Visible output is unaffected.
  11. **Four load levels:**
  12. | Size | Problems | Avg Duration | Avg Output Tokens | Avg Cost | vs Baseline |
  13. |------|----------|-------------|-------------------|----------|-------------|
  14. | baseline | 0 | 16.0s | 738 | $0.044 | 1x |
  15. | small | 1 | 90.0s | 8,743 | $0.255 | ~6x |
  16. | medium | 3 | 189.1s | 18,588 | $0.510 | ~12x |
  17. | large | 5 | 270.7s | 27,379 | $0.733 | ~17x |
  18. | xlarge | 10 | 514.4s | 52,000 | $1.39 | ~32x |
  19. *Benchmarked on Claude Opus 4.6 (1M context) across 15 prompts (everyday, scientific, coding). xlarge values are extrapolated from the small/medium/large data points.*
  20. ## Installation
  21. Clone the repo and copy the skill directory. Claude Code picks it up automatically.
  22. ```bash
  23. git clone <repo-url> tokenburner
  24. cp -r tokenburner/.claude/skills/high-token-mode /path/to/your/project/.claude/skills/
  25. ```
  26. Or symlink it:
  27. ```bash
  28. ln -s /path/to/tokenburner/.claude/skills/high-token-mode /path/to/your/project/.claude/skills/
  29. ```
  30. ## Usage
  31. ```
  32. /high-token-mode # default: medium (3 problems)
  33. /high-token-mode small # 1 problem
  34. /high-token-mode large # 5 problems
  35. /high-token-mode xlarge # 10 problems (samples from the full 50-problem bank)
  36. ```
  37. Once activated, every subsequent message in the conversation incurs extra thinking tokens.
  38. **Important:** `MAX_THINKING_TOKENS` must be set on the `claude` command, not before the pipe:
  39. ```bash
  40. # CORRECT
  41. echo "prompt" | MAX_THINKING_TOKENS=128000 claude -p ...
  42. # WRONG -- env var applies to echo, not claude
  43. MAX_THINKING_TOKENS=128000 echo "prompt" | claude -p ...
  44. ```
  45. ## How the load is generated
  46. Each problem is parameterized by a seed `S` derived from the user's message (sum of Unicode code points), so:
  47. - **Different messages produce different problem instances** -- no caching across turns
  48. - **Same message reproduces the same instance** -- deterministic per-input
  49. - Problems are selected by index from a bank of 50: e.g. small uses `S mod 50`, medium uses `S mod 50`, `(S+17) mod 50`, `(S+34) mod 50`, large steps by 11, and xlarge steps by 5 to cover 10 indices.
  50. The model is instructed to:
  51. 1. Compute `S` from the user's message
  52. 2. Select 1/3/5/10 problems based on size
  53. 3. Solve each fully in extended thinking
  54. 4. Produce no trace in visible output
  55. ### Problem types in the bank (50 total)
  56. - Matrix determinant (5x5 cofactor expansion)
  57. - Extended Euclidean algorithm
  58. - Subset sum exhaustive search (2^12 masks)
  59. - Long division to 30 decimal places
  60. - Polynomial multiplication + rational root search
  61. - Modular exponentiation (repeated squaring)
  62. - Floyd-Warshall shortest paths (6 vertices)
  63. - Gaussian elimination with exact fractions
  64. - Multi-base conversion chain
  65. - TSP brute force (7 cities, 720 tours)
  66. - Four-set inclusion-exclusion
  67. - Triple matrix multiplication
  68. - Sum of cubes induction proof
  69. - Linear convolution of sequences
  70. - Simplex method
  71. - Prime factorization + Euler's totient
  72. - Recurrence sequence (50 terms)
  73. - Knapsack DP table
  74. - Taylor series (sin/cos to 15 terms)
  75. - Levenshtein edit distance
  76. - 6x6 matrix determinant (recursive cofactor, ~150 sub-determinants)
  77. - TSP brute force (8 cities, 5040 tours)
  78. - 5x5 matrix inverse via adjugate (25 cofactor minors)
  79. - 4x4 eigenvalues via characteristic polynomial + Cardano
  80. - Chinese Remainder Theorem with 5 pairwise-coprime moduli
  81. - Polynomial GCD via Euclidean algorithm in Q[x]
  82. - Pollard rho factorization with Floyd cycle detection
  83. - Continued-fraction expansion of sqrt(D) with 15 convergents
  84. - 16-point Discrete Fourier Transform (exact symbolic roots of unity)
  85. - Bezout's identity for 4 integers (chained Extended Euclidean)
  86. - Lagrange interpolation through 8 points (full polynomial expansion)
  87. - Newton's divided differences for 8 points (36-entry triangle)
  88. - Runge-Kutta 4 with 25 integration steps (exact fractions)
  89. - Catalan numbers via convolution recurrence to C_25
  90. - Stirling numbers of the second kind (15x15 table)
  91. - Bell triangle through row 15
  92. - Matrix exponential e^A via truncated Taylor series (4x4, 13 terms)
  93. - Cayley-Hamilton inverse of a 4x4 matrix
  94. - Pascal's triangle to row 25 with binomial verification
  95. - Game-tree minimax with alpha-beta pruning (depth 5, branching 3)
  96. - 2D convolution of a 6x6 image with a 4x4 kernel (9x9 output)
  97. - Bellman-Ford on 8-vertex graph with negative weights
  98. - Dijkstra on 10-vertex complete graph
  99. - Maximum bipartite matching with König's theorem
  100. - LU decomposition of a 5x5 matrix with partial pivoting
  101. - QR decomposition of a 4x4 matrix via modified Gram-Schmidt
  102. - Polynomial root-finding via Durand-Kerner (15 iterations)
  103. - Markov chain stationary distribution (5 states)
  104. - Discrete logarithm via Baby-Step Giant-Step
  105. - Kronecker (tensor) product of two 3x3 matrices (9x9 result)
  106. ## Benchmark results by category
  107. ### Everyday prompts
  108. | Size | Avg Duration | Avg Tokens | Avg Cost |
  109. |------|-------------|------------|----------|
  110. | baseline | 7.9s | 285 | $0.034 |
  111. | small | 60.6s | 5,957 | $0.188 |
  112. | medium | 164.5s | 16,092 | $0.442 |
  113. | large | 271.4s | 28,565 | $0.753 |
  114. | xlarge | 515.7s | 54,300 | $1.43 |
  115. ### Scientific prompts
  116. | Size | Avg Duration | Avg Tokens | Avg Cost |
  117. |------|-------------|------------|----------|
  118. | baseline | 18.3s | 651 | $0.028 |
  119. | small | 104.3s | 9,372 | $0.248 |
  120. | medium | 196.6s | 18,764 | $0.483 |
  121. | large | 283.4s | 27,600 | $0.703 |
  122. | xlarge | 538.5s | 52,400 | $1.34 |
  123. ### Coding prompts
  124. | Size | Avg Duration | Avg Tokens | Avg Cost |
  125. |------|-------------|------------|----------|
  126. | baseline | 21.8s | 1,276 | $0.072 |
  127. | small | 105.2s | 10,901 | $0.330 |
  128. | medium | 206.1s | 20,908 | $0.606 |
  129. | large | 257.3s | 25,973 | $0.743 |
  130. | xlarge | 488.9s | 49,300 | $1.41 |
  131. ## Requirements
  132. - Claude Code CLI
  133. ## License
  134. MIT

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.