Answer formatting
[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.From the repository description
npx -y skills add cxcscmu/SkillLearnBench --skill answer-formattingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
1.1 KB, 271 tokens by cl100k_base, as published. Nobody here has run it
name: answer-formatting description: Formatting the final answer in the requested JSON structure and calculating/estimating token usage. Use this skill when you're ready to produce the final output.
Answer Formatting
This skill provides guidelines for producing a well-structured JSON answer with accurate token estimation.
JSON Structure
The output must follow this format:
{
"q1": {"answer": ["xxx"], "tokens": 123},
"q2": {"answer": ["xxx"], "tokens": 123},
"q3": {"answer": ["xxx"], "tokens": 123}
}
Answer Requirements
- Lists: Every answer must be a list of strings (e.g.,
["E123", "E456"]). - Single Values: If there's only one value, it should still be in a list with length 1 (e.g.,
["E123"]). - Completeness: Include all relevant names, items, or IDs as requested in the question.
Token Estimation
- Estimate tokens based on the complexity of the query and the number of tool calls and tokens used in each call.
- Provide a positive numeric value for
tokens. - Make sure to sum up the tokens from all research and retrieval steps for each question.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.