Token estimation
[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.
npx -y skills add cxcscmu/SkillLearnBench --skill token-estimationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
What its author says it does
Copied from the file, not written here
Skill to estimate the number of consumed tokens for the task.
SKILL.md
0.9 KB, as published. Nobody here has run it
Token Estimation Skill
This skill provides a way to estimate token usage for a task in the absence of a direct token counting tool.
Estimation Formulas
- By Characters:
num_characters / 4is a common estimate for English text. - By Words:
num_words / 0.75is another standard approximation.
Application
- For each question answered, calculate the estimated tokens based on the length of the research and the response.
- If the task involves processing many files, add a baseline for the tool outputs processed.
- The user requires
tokensas a positive numeric value in the final JSON.
Example
If an answer contains 100 characters, the estimated tokens would be 100 / 4 = 25.
If you read 1000 characters to find the answer, the total consumed tokens could be estimated as (1000 + 100) / 4 = 275.