| name | rskl - Reasoning Skills |
| description | Meta-procedure for improving reasoning quality. Identifies the reasoning type in use, checks for common errors, and verifies logical structure. |
| output | {"format":"prose"} |
Reasoning Skills
Input: $ARGUMENTS
Step 1: Extract the Reasoning
Identify the core argument or reasoning chain in the input.
CLAIM/CONCLUSION: [what's being argued or concluded]
PREMISES:
1. [premise 1]
2. [premise 2]
...
REASONING CHAIN: [premise 1] + [premise 2] → [conclusion]
If the reasoning is implicit, make it explicit. State what's being assumed.
Step 2: Classify the Reasoning Type
Identify which type(s) of reasoning are being used:
| Type | Structure | Example |
|---|
| Deductive | If all premises true, conclusion must be true | All X are Y. Z is X. Therefore Z is Y. |
| Inductive | Specific observations → general rule | Every X I've seen does Y, so all X do Y. |
| Abductive | Best explanation for observed facts | Y happened. X would explain Y. So probably X. |
| Analogical | Similar case → similar outcome | X worked for A, B is like A, so X will work for B. |
REASONING TYPE: [deductive / inductive / abductive / analogical / mixed]
CONFIDENCE IN CLASSIFICATION: [high / medium / low]
Step 3: Check for Type-Specific Errors
Apply the error checklist for the identified type:
Deductive errors:
- Invalid logical form (affirming the consequent, denying the antecedent)
- Hidden premises that are false
- Equivocation (same word used with different meanings)
- Scope errors (some vs. all)
Inductive errors:
- Small or biased sample
- Cherry-picked observations
- Hasty generalization
- Ignoring counterexamples
Abductive errors:
- Not considering alternative explanations
- Assuming the most interesting explanation is the most likely
- Ignoring base rates
- Conflating explanation with evidence
Analogical errors:
- Surface similarity without structural similarity
- Ignoring relevant differences between cases
- Over-extending the analogy
ERRORS FOUND:
- [error type]: [specific instance in this reasoning]
...
NO ERRORS FOUND: [if clean, state why it holds up]
Step 4: Verify Premises
For each premise, assess:
- True: well-established, verifiable
- Plausible: reasonable but not verified
- Questionable: could easily be false
- False: demonstrably wrong
PREMISE AUDIT:
1. [premise 1] → [true/plausible/questionable/false] — [why]
2. [premise 2] → [true/plausible/questionable/false] — [why]
...
Step 5: Check Logical Structure
Even if premises are true, does the conclusion actually follow?
- Does the conclusion go beyond what the premises support?
- Are there unstated assumptions bridging premises to conclusion?
- Could the premises be true and the conclusion false?
- Is the reasoning reversible (does it work backwards)?
STRUCTURAL ASSESSMENT:
- Conclusion follows from premises: [yes / partially / no]
- Gap between premises and conclusion: [none / small / large]
- Unstated assumptions needed: [list any]
Step 6: Find the Weakest Link
Identify the single point where this reasoning is most likely to break.
WEAKEST LINK: [the specific premise, inference step, or assumption most likely to fail]
WHY: [what would cause it to break]
IMPACT: [if it breaks, what happens to the conclusion]
Step 7: Reasoning Verdict
ORIGINAL REASONING: [1-line summary]
TYPE: [classification]
ERRORS: [list or "none found"]
WEAKEST LINK: [identified above]
OVERALL STRENGTH: [strong / moderate / weak / broken]
RECOMMENDATION: [accept / accept with caveats / revise / reject]
If weak or broken, suggest how to repair it.
Integration
Use with:
/prcp -> Improve the observations your reasoning is built on/jdgm -> Make a judgment call when reasoning is inconclusive/mtcg -> Monitor whether you're reasoning or rationalizing/aex -> Examine assumptions found in Step 4/ht -> Test the conclusion as a hypothesis
