agents2agentsai/conversation-report

Conversation Report

Generate a cleaned-up, organized version of what was actually discussed in this conversation. This is NOT a generic survey — it reflects the user's specific exploration.

For exhaustive deep reports from KB cards, use $cross-paper-report.

Core Principle: Clean Up, Don't Regenerate

Reuse phrasing and explanations from the conversation — fix conversational artifacts (hedging, self-correction, tangents) into clean prose. Organize by question/topic, not chronologically or by paper. Add light connective tissue but do not pad with new analysis.

Phase 1: Understand the Conversation

Figure out: what was the implicit research question? What was the exploration path? Which papers, comparisons, and insights emerged? What did the user care most about?

Phase 2: Generate the Report

Organize by what the user explored. Structure: The Question (what drove the exploration), What We Found (cleaned-up explanations organized by topic, not by paper), Key Insights (3-5 "aha moments"), Papers Referenced (table with role in conversation), Open Questions (actionable items with suggested $paper-discovery or $paper-synthesis follow-ups).

Present via present_reading_view. Each section ≤ 30 lines. Include relevant figures via crop_and_store_figure where they aid understanding.

Phase 3: Persist

Four outputs in order:

1. Chat response (mandatory) — the reading view is the deliverable.
2. KB card updates (skip if KB disabled) — fire-and-forget $kb subagent for each card where discussion produced new insights. Do not ask the user.
3. Journal entry (skip if KB disabled) — prepend to <kb_path>/research-journal.md: date, topics explored, conclusions, open questions, cards touched. Keep under 20 lines. Read existing content first, write new entry + existing back.
4. Research context (skip if KB disabled) — if conversation revealed user preferences or project context, offer to update <kb_path>/research-context.md.

Graceful Degradation

Short conversation: still produce the report, note brevity. No KB cards: still report, suggest $paper-synthesis. No KB configured: chat-only output. Unfocused conversation: organize into separated sections.