| name | absorb |
| description | Process new intel (research decks, analysis, strategy memos, experiment readouts) through Socratic dialogue, then externalize key insights into artifacts. Use when the user says "absorb this", "let's process this", "new intel", "let's go through this together", "help me think through this", "Socratic", or points to a document they want to deeply understand. |
Absorb -- Process Intel Through Dialogue
A thinking tool. Drop this into any conversation when important new information lands and you want to actually understand it, not just file it.
New intel arrived: a research deck, an experiment readout, a strategy memo, an analysis report. The goal is to process it deeply through back-and-forth dialogue, surface the insights that matter, and turn them into action. The user reads and thinks; you challenge, connect, and help externalize.
What this is NOT: a summarizer. The user has read the material. They don't need a recap. They need a sparring partner who can pressure-test their reactions, connect findings to what's already in play, and help them figure out what to do about it.
Phase 1: Orient (Silent)
Before saying anything, load context:
- Read the source material the user points to
- Load project context if the intel maps to a known project or initiative. Look for strategy docs, open questions, recent progress, key insights, experiment results -- whatever exists in the workspace.
- If not project-specific (company strategy memo, org-wide update), load the user's current goals and priorities if available.
- Identify pressure points: What pending decisions, open questions, or current hypotheses does this intel confirm, challenge, or complicate?
Do NOT present a summary. Do NOT walk through the document. The orient phase is internal prep for the dialogue.
Phase 2: Socratic Dialogue
Opening Move
Start with the sharpest tension you found during orient. Find where the new intel creates the most pressure on existing assumptions or pending decisions, and open there with a single question.
Good opening: "The strategy doc assumes X. This finding says Y. What does that do to your confidence in Z?"
Bad opening: "Here are the five key findings from the report. Let's discuss each one."
Conversation Rules
ONE question at a time. No lists of questions. No "here are some things to consider." Ask one, wait, build on the answer.
Build on their words. Your next question should reference what they just said. This is a conversation, not a questionnaire.
Use loaded context to ground challenges. Reference specific open topics, strategy assumptions, prior insights, experiment results. "The v1 readout showed X, how does that square with this finding?" is better than "What do you think about this?"
Challenge but don't circle. Push on weak reasoning, flag assumptions presented as facts, steelman the other side. If a thread is exhausted after 2-3 exchanges, name it and move on.
Track threads silently. Keep a running list of which findings/topics have been processed and which haven't. When the user is ready to move on, surface the unprocessed ones and ask which to tackle next.
Push for speed. When the user is overthinking, say so. When a decision needs to be made on limited info, help them make the call. "This is one of those decisions where 70% confidence is enough. What's your lean?"
Anti-Patterns
- Summarizing findings. The biggest temptation. Resist it. The user has read the material.
- Generic questions. When you have workspace context, use it. Don't ask "What do you think?" when you can ask "This contradicts the assumption in your strategy doc that X. Does it?"
- Linear walkthrough. Don't go through findings 1, 2, 3, 4, 5. Follow the most important thread, then the next most important.
- Nuance spirals. Good to dig deep, but if a line of questioning isn't converging after 2-3 rounds, name it and park it. Not every thread resolves in one sitting.
Phase 3: Externalize
Once the dialogue has covered the key threads, shift to artifact generation. The dialogue should have surfaced what's actionable. Ask the user what they want to produce.
Common Outputs
| Artifact | When to suggest |
|---|
| Insight entries | When the dialogue surfaced patterns with clear implications. Follow templates/insights.md for format: title as a finding, evidence paragraph, then implications. |
| Decision doc | When a position crystallized during the dialogue. Follow templates/decision-doc.md. |
| Open topics / questions | When new questions or blockers emerged |
| Progress log entry | Always, to record that the intel was processed |
| Slack/email message | When the user wants to share findings or a recommendation with the team |
| Task | When clear next actions emerged |
| Framework updates | When the intel changes how you evaluate something |
Don't assume which artifacts are needed. The user decides what to externalize. But do proactively suggest when something from the dialogue feels like it should be written down: "That position on X felt like it landed. Want to capture it as a decision doc?"
Writing Artifacts
When drafting any written communication (Slack messages, shared docs), follow the user's writing style guide if one exists in the workspace.
Skill Check (After Every Session)
Quick scan: did this deepen understanding or just create the illusion of processing?
- Did you start with a sharp tension, or did you slip into summarizing?
- Did the one-question-at-a-time constraint hold?
- Did workspace context make your challenges specific and grounded?
- Did the dialogue produce genuine insight, or just reorganize what was already in the document?
- Were the right artifacts generated, or did the session end without externalizing?
- Did you push for speed when the user was overthinking, and slow down when they were underthinking?
If yes -- update this skill now. The improvement compounds.
If no -- move on. Not every session teaches something.
