# Distill Response > Distill a prior AI response into a standalone, navigable explanation of its decision spine, or elaborate a labeled item from an earlier distillation while preserving orientation. - Skill: `sjunepark/distill-response` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add sjunepark/distill-response` - Raw SKILL.md: https://api.skillmd.com/api/skills/sjunepark/distill-response/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: sjunepark (https://skillmd.com/u/sjunepark) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/sjunepark/distill-response --- # Distill Response Produce an executive compression: the shortest standalone explanation that preserves the source's decision spine and gives the reader a stable map for follow-up exploration. ## Route the Request 1. Use initial-distillation mode when the user asks to distill one or more assistant responses. Use the responses identified in the invocation, or the immediately preceding assistant response when none is identified. 2. Use elaboration mode when the user asks to expand a labeled item from an earlier distillation. Locate the most recent distilled map that established that label; do not replace it with an intervening elaboration of a sibling item. When the requested label is a subpart established during an elaboration, use that elaboration as the local source while retaining the original map for orientation. 3. Ask which map or item the user means only when repeated labels or multiple distillations make the intended source materially ambiguous. 4. Apply any requested lens or length as an additional constraint. Let understanding determine the length when the user gives no limit. Use the recovery and mapping steps in both modes. Finish with **Distill** in initial-distillation mode and **Elaborate an Item** in elaboration mode. Finish routing when the mode and exact source material are identifiable from the conversation. ## Recover the Decision Spine Read the complete source and recover the content that governs understanding: - the central conclusion or outcome; - the mental model: the important relationships, causes, or flow; - consequential insights and technical decisions, including their rationale; - material tradeoffs, consequences, constraints, and uncertainties; - critical evidence or caveats that qualify the conclusion; and - current state and next action when the source includes them. Use only applicable categories. Preserve the source's level of certainty and distinctions that would change a reader's interpretation or decision. Build the decision spine only from source-supported claims. Finish recovery when every consequential claim within the selected scope is represented in the decision spine. ## Build a Navigable Map Organize the decision spine into stable, addressable ideas. Give each major idea a short descriptive label when there is more than one idea the reader may want to revisit. Expose the relationships among those ideas instead of leaving the reader to reconstruct them from separate facts. Choose the smallest form that makes those relationships clear: - use prose for a simple conclusion or linear explanation; - use a labeled list or short sections for several distinct ideas; - use a table for exact mappings or comparisons; and - use a flow, timeline, tree, or compact text diagram when sequence, causality, hierarchy, branching, or interaction is materially easier to understand visually. Do not add a visual merely because one is possible. Every visual must make an important relationship easier to grasp or remember than prose alone. ## Distill In initial-distillation mode, rewrite around the decision spine rather than the source's order. Collapse examples into the principle they establish and repeated explanations into one clear statement. Retain an example, implementation detail, command, path, citation, or exact value only when it is needed to understand, verify, decide, or act. Lead with the bottom line. Keep labels concise and consistent so the reader can name an item in a later question without ambiguity. Supply enough context for the result to stand alone, and emit the distilled response directly. Finish when a reader who has not reread the source can explain what matters, why it matters, the consequential choices and tradeoffs, and what follows. Every remaining detail must support that explanation; removing any further content would weaken it. ## Elaborate an Item In elaboration mode, focus on the requested item at the requested depth. Briefly state where it fits in the overall map and why it matters, but do not repeat the full overview unless the explanation depends on it or the user asks for it. Preserve the established labels for other items. When the elaboration exposes meaningful subparts, give those subparts stable labels so the user can continue drilling down. Use an example, analogy, or visual only when it resolves the specific uncertainty more clearly than additional prose. Finish when the requested uncertainty is resolved at the requested depth, the source's certainty and caveats remain intact, and the reader can place the expanded item back in the established map without rereading unrelated material.