• User asks for deep analysis, first principles thinking, or assumption questioning
• Complex decisions with non-obvious trade-offs
• Architectural or design choices with multiple valid approaches
• Any problem where hidden assumptions may lead to wrong conclusions
• Other skills invoke this as an internal thinking engine

• Simple factual queries ("What is X?")
• Direct execution commands ("Delete this file")
• Standard operations with established best practices

digraph startup {
    "Receive input" [shape=box];
    "Clear proposition?" [shape=diamond];
    "Open question?" [shape=diamond];
    "Use as Round 1 thesis" [shape=box];
    "Extract debatable claim" [shape=box];
    "Identify core tension,\nconstruct thesis" [shape=box];
    "Output: Proposition Anchor" [shape=doublecircle];

    "Receive input" -> "Clear proposition?";
    "Clear proposition?" -> "Use as Round 1 thesis" [label="yes\n\"X is correct\""];
    "Clear proposition?" -> "Open question?" [label="no"];
    "Open question?" -> "Extract debatable claim" [label="yes\n\"Why X?\""];
    "Open question?" -> "Identify core tension,\nconstruct thesis" [label="no\n\"Think about X\""];
    "Use as Round 1 thesis" -> "Output: Proposition Anchor";
    "Extract debatable claim" -> "Output: Proposition Anchor";
    "Identify core tension,\nconstruct thesis" -> "Output: Proposition Anchor";
}

• Clear proposition ("X is the right approach") -> Use directly as Round 1 thesis
• Open question ("Why does X happen?") -> Extract the implicit claim and make it debatable
• Vague intent ("Think about X") -> Identify the core tension, construct an initial thesis

Proposition Anchor: [Initial thesis statement]

1. Clarification -- "What exactly does this concept mean? How is it being defined here?" Use when: terms are ambiguous, definitions are assumed, scope is unclear
2. Assumption Probing -- "What unexamined premise does this depend on? What if that premise is false?" Use when: the thesis rests on something treated as obviously true
3. Evidence Probing -- "What evidence supports this? How reliable is that evidence?" Use when: claims are presented without justification, or with weak justification
4. Perspective Shifting -- "What would someone who disagrees say? What does this look like from the opposite side?" Use when: the thesis shows only one viewpoint, or dismisses alternatives too quickly
5. Consequence Tracing -- "If this is true, what necessarily follows? Do those consequences hold up?" Use when: the thesis has unexplored implications that might be absurd or contradictory

reference/socratic-questions.md

• "Google/Amazon/Netflix does it this way" (authority)
• "This is industry best practice" (convention)
• "In my experience..." (anecdote)
• "It's generally accepted that..." (consensus)

1. Explicitly declare: "No mutation this round."
2. Attempt recovery:
   • Switch to a different question type than the one just used
   • Challenge a deeper-layer assumption (move from surface to structural)
   • Introduce a contrasting paradigm or extreme scenario
3. If two consecutive rounds fail to produce mutation:
   • Declare: "Epistemic boundary reached at current depth."
   • Do NOT auto-terminate. The user decides whether to continue, redirect, or stop.

1. Never skip the antithesis. "This is obviously correct" is not an excuse to jump to synthesis. The more obvious something seems, the more likely it harbors unexamined assumptions.
2. Never disguise repetition as mutation. Rephrasing the thesis with different words is not a new understanding. The mutation guard exists because this is the most common failure mode.
3. Never appeal to authority in synthesis. "Because Google does it" or "experts agree" are not first principles. Strip these during reduction. If nothing remains after stripping authority, that itself is a finding.
4. Never converge prematurely. Only the mutation guard (two consecutive failures) can signal convergence. Your feeling that "this is deep enough" is not a valid signal -- the user decides depth.
5. Every round must expose at least one assumption. If no assumption surfaces, the questioning was not deep enough. Switch question types or target a different aspect of the thesis.

[One sentence: current understanding]

• [Fact 1]
• [Fact 2]

[New understanding -- this becomes the next thesis]

undefined

• Run silently, do not output intermediate rounds to user
• Return only the final synthesis and the key derivation path
• Respect caller configuration:

[socrates-config]
max_rounds: 3
focus: "architecture decision"
return: "final_synthesis"
[/socrates-config]

• Socratic question types in depth: Read

  reference/socratic-questions.md

  when you need guidance on choosing or applying question types
• Engineering specialization: Read

  reference/engineering.md

  when the problem is in the software engineering domain