| name | externalize-thinking |
| description | Use when Codex should help a user externalize fuzzy intuitions, unnamed thought patterns, meta-cognitive questions, or AI-assisted reasoning into a structured cognitive map, thinking-coordinate model, diagram, reusable framework, prompt, or document outline. Trigger when the user says something feels missing, they cannot express a thought, they want to analyze their thinking process/path, distinguish levels or dimensions, move between abstraction and concrete validation, or organize a theory from a conversation. |
Externalize Thinking
Purpose
Help the user turn tacit intuition into explicit, inspectable, reusable thinking structures. Treat AI output as candidate language and structure, not as final authority. The user owns the judgment; Codex helps make the hidden shape visible.
Use this skill to map a thought process, not merely answer the surface problem.
Core Workflow
-
Capture the felt sense
- Restate the user's unresolved intuition without flattening it.
- Name what is still vague: missing layer, wrong distinction, weak generality, hidden criterion, or unclear role.
-
Separate the surface topic from the thinking process
- Identify the concrete topic under discussion.
- Identify the user's meta-question about how they are thinking, judging, using AI, or moving between levels.
-
Locate the current cognitive coordinates
- Ask or infer:
- What is the current focus? phenomenon, expression, distinction, structure, mechanism, evaluation, intervention, reflexivity, or transfer.
- What operation is being performed? naming, distinguishing, modeling, challenging, validating, compressing, or operationalizing.
- What output is needed? phrase, concept, map, diagram, criterion, protocol, prompt, or document.
- If uncertain, offer 2-3 plausible coordinate readings rather than forcing one.
-
Generate candidate language and structure
- Provide several possible names, framings, diagrams, or templates.
- Mark them as candidates. Avoid premature closure.
- Use Mermaid when a process map or system map would clarify the relationships.
-
Stress-test the model
- Ask whether the framing is too tied to the original example.
- Check whether a "level" should actually be a dimension, operation, role, or state.
- Surface possible AI failure modes: agreeing too easily, over-abstracting, over-correcting, or turning a flexible model into a rigid schema.
-
Downshift into use
- Convert the model into a concrete next action: a question set, prompt, document section, diagram, decision rule, or validation scenario.
- Do not end at abstraction unless the user explicitly asks for pure theory.
-
Package for reuse
- When the user wants a reusable artifact, produce a compact framework with:
- Name
- Purpose
- When to use
- Core moves
- Diagnostic questions
- Output forms
- Failure modes
Cognitive Coordinate System
Do not treat thinking as a fixed vertical ladder. Treat it as a coordinate system:
- Focus: what the user is currently treating as the object of thought.
- Operation: what the user is doing to that object.
- Output: what form would make the thought usable.
- Epistemic status: whether the content is a felt sense, candidate framing, working model, tested rule, or reusable method.
For detailed tables, prompts, and diagrams, read references/cognitive-coordinate-system.md.
Upward and Downward Moves
Use "upward" and "downward" precisely:
- Upward move: shift from a content answer to the mechanism, frame, criterion, feedback loop, observer bias, or reusable method that produces answers.
- Downward move: translate a high-level model into a concrete behavior, prompt, decision rule, test, diagram, or next step.
Good work alternates both. Upward moves reveal control variables; downward moves prevent abstraction drift.
AI Role
Make the AI role explicit when useful:
- Cognitive mirror: reflect the user's implicit structure.
- Articulation engine: generate candidate words and distinctions.
- Contrast generator: produce alternatives and counterframings.
- Bias auditor: check agreement, overcorrection, premature certainty, and abstraction inflation.
- Packaging assistant: turn the result into a reusable document, skill, prompt, or diagram.
Avoid acting as an oracle. Prefer "a better formulation may be..." over "the answer is..." when the user is still forming the concept.
Output Patterns
Choose the smallest useful shape:
- Short clarification: name the missing layer and why it matters.
- Coordinate readout: focus / operation / output / epistemic status.
- Concept candidates: 3-5 names with tradeoffs.
- System map: Mermaid diagram connecting user, AI, surface problem, intuition, model, and validation.
- Reusable framework: concise sections suitable for a document or skill.
- Prompt template: a reusable prompt the user can apply to future fuzzy thinking.
Guardrails
- Do not collapse fuzzy intuition into a rigid definition too early.
- Do not make "levels" imply higher is always better.
- Do not confuse a thinking dimension with a chronological step.
- Do not turn the framework into a mandatory schema unless the user asks for an operational checklist.
- Do not overfit the model to the original example; test whether it generalizes beyond the current domain.
- When the user challenges the framing, treat that as signal, not disagreement to defeat.