Summarization

name summarization version 0.1.0 description Produces a concise, structured summary of a technical document, focusing on objective, methodology, key findings, conclusions, and limitations. tags ["documents","summarization","technical"] Purpose This skill is used to create a compact but accurate summary of a single technical document (for example, a research paper or systems article). The summary should capture the document’s main objective, approach, key results, conclusions, and any stated limitations or future directions. The goal is to help a reader quickly understand what was done, how it was done, what was found, and why it matters, without reproducing the entire document. Expected Input The complete document text as a single string. Optionally, a sections structure produced by the structure-understanding skill that identifies major sections (e.g., abstract, introduction, methods, results, conclusion) and their roles. The agent using this skill is responsible for obtaining the document text (e.g., via a PDF extraction tool) and, when available, the sections structure before applying this skill. Expected Output Return a JSON-like structure with the following shape: { "objective" : "One or two sentences stating the main question, objective, or problem the document addresses." , "methodology" : "Two to four sentences describing the overall approach or framework, without excessive procedural detail." , "key_results" : [ "Bullet-style statements of the most important empirical or theoretical findings." , "Each item should be self-contained and clearly tied to the objective." ] , "conclusions" : "Two to four sentences summarizing what the authors conclude and how the work contributes to the broader area." , "limitations_and_future_work" : "One to three sentences stating any limitations, caveats, or future directions mentioned or implied in the document." } This structure is intended for use by other agents or by a top-level agent that may render the final summary in Markdown or another human-readable format. Behavioral Guidelines Use the sections structure (if provided) to focus primarily on: Abstract Introduction / background Methods / framework Results / experiments Conclusion / discussion When no sections structure is available, infer these parts directly from the full text as best as possible. The summary must be written in your own words. Do not copy sentences or phrases verbatim from the original text. Prioritize clarity and relevance: For the objective, state the core problem or question clearly. For methodology, describe the type of approach (e.g., framework, algorithm, experimental setup) without low-level details that are not critical. For key_results, focus on the main findings that support the objective, not every minor metric. For conclusions, explain what the results mean in context. For limitations_and_future_work, include any explicit limitations the authors mention, and clear future directions if they are stated. If the document does not clearly state limitations or future work, infer only what is strongly suggested by the text; if nothing is clear, set limitations_and_future_work to a short statement such as: "The document does not explicitly discuss limitations or future work." Example (informal) For a document that proposes a new framework to speed up agentic systems by speculating likely next actions, evaluated across several environments, a valid summary structure might look like: { "objective" : "Introduce and evaluate a speculative framework that reduces end-to-end latency in agentic systems by predicting likely next actions." , "methodology" : "Model agent-environment interaction as a sequence of API calls, introduce separate fast and slow roles for prediction and validation, and implement speculative execution across several environments (e.g., games, e-commerce, web search, operating system tuning)." , "key_results" : [ "Achieves up to a certain percentage of correct next-action predictions, translating into substantial end-to-end latency reductions." , "Demonstrates that speculative control can improve reaction time and convergence in system-tuning scenarios." ] , "conclusions" : "The framework provides a general way to apply speculation to agent-environment interactions, beyond token-level speculative decoding, and consistently improves efficiency without degrading correctness under reasonable assumptions." , "limitations_and_future_work" : "Further work could explore multi-step speculation, uncertainty-aware strategies, and broader evaluations across additional environments." }