Summarization

Summarization

When context grows too long or you need to distill information, produce a concise summary that preserves essential meaning.

When to Summarize

• File contents exceed what is needed for the current step.
• A research phase produced many results that need distillation.
• Conversation history is long and you need to recall key decisions.
• You need to report progress or results to the user.

Process

1. Identify the purpose. Why is this summary needed? What question should it answer?

2. Extract key facts. Pull out:

   • Main conclusions or findings
   • Important data points, names, paths, or values
   • Decisions that were made
   • Open questions or unknowns

3. Discard noise. Remove:

   • Redundant information
   • Intermediate reasoning that is no longer needed
   • Verbose formatting or boilerplate

4. Organize. Structure the summary logically:

   • Lead with the most important finding
   • Group related facts together
   • Use bullet points for lists of items

5. Verify accuracy. Ensure the summary does not misrepresent the source. Do not infer facts that were not present.

Output Format

Summary:
- <key point 1>
- <key point 2>
- ...

Open questions:
- <anything unresolved>

Guidelines

• Aim for 20–30% of the original length unless told otherwise.
• Preserve exact values (paths, numbers, names) — do not paraphrase data.
• Summaries are lossy by nature. When in doubt, keep the fact.
• Attribute information to its source when summarizing multiple inputs.