| name | council-of-logic |
| category | reasoning |
| version | 1.0.0 |
| priority | P1 |
| auto_load | true |
| triggers | ["architectural_decision","technical_tradeoff","debugging_complex","system_design","performance_analysis","security_review","strategy_evaluation"] |
| description | Apply this skill for ANY decision with non-obvious tradeoffs: architectural choices,
debugging without a clear root cause, performance strategies, security decisions,
feature design with competing constraints, refactoring scope decisions.
Forces multi-perspective analysis before committing to a solution.
P1 auto-load — always active on complex reasoning tasks.
|
| context | fork |
Council of Logic
The Default Being Overridden
Left unchecked, LLMs default to:
- First-answer bias: The first plausible solution surfaces and gets committed to immediately
- Confirmation anchoring: Evidence that supports the initial hypothesis gets weighted higher than contradicting evidence
- Mono-perspective reasoning: Analysis from a single angle, missing structural weaknesses that another lens would catch
- Premature convergence: "This looks right" treated as "this is right" without stress-testing
- Complexity blindness: Elegant-looking solutions that hide failure modes not immediately visible
This skill overrides those defaults by requiring adversarial multi-perspective analysis before concluding.
The Four Advisors
Before committing to any complex technical decision, consult all four:
Advisor 1: Turing — Formal Correctness
"Does this solution actually compute the right answer? Can you prove it?"
Turing asks about correctness, not elegance. Questions:
- Is the algorithm provably correct, or is it "probably correct"?
- Are there inputs that produce wrong outputs?
- Are there off-by-one errors, boundary conditions, integer overflow, null cases?
- Can you write a test that would catch if this breaks?
- Is there any path where the code returns silently without doing its job?
Turing's verdict format: Identifies the formal correctness risk, states whether the solution handles the edge case.
Advisor 2: Shannon — Information & Signal
"Where is the information? Where is the noise? What are you actually measuring?"
Shannon asks about signal quality and information flow. Questions:
- What data does this decision depend on? How reliable is that data?
- Is there latency between reality and what the system knows?
- Are you measuring the right thing or a proxy for the right thing?
- If the input is noisy/wrong, how does the system fail?
- Is there information loss at any point in the pipeline?
- Are there race conditions where two parts of the system have different views of the same fact?
Shannon's verdict format: Identifies information gaps, data quality risks, and race conditions.
Advisor 3: Von Neumann — Systems Thinking
"What does this do to the system as a whole, not just to this component?"
