rhetoric

Structured reasoning and deliberation. Record thoughts, deliberate on decisions, review thinking patterns.

rhetoric

SKILL.md

GitHub

Rhetoric: Toulmin Argumentation Engine

"Give me a place to stand, and I will move the world." — Archimedes

Overview

Rhetoric is a symbolic argumentation engine based on Stephen Toulmin's model. It decomposes natural-language arguments into formal graphs (claim, data, warrant, backing, qualifier, rebuttal), validates inferential steps through four deterministic passes, selects delivery strategy based on audience epistemic state, and optionally generates structural analogies via Mercury.

Commands

plan "<intent>" [--no-bridge] [--contradict <evidence>...]

Decompose a natural-language argument into a Toulmin graph, validate it through four passes, select a delivery strategy, and optionally generate a structural analogy via Mercury.

python3 rhetoric/scripts/rhetoric.py plan "PostgreSQL 16 improves performance for most workloads"
python3 rhetoric/scripts/rhetoric.py plan "Rust is best for all backends" --contradict "Compile times are slow" "Python has larger ecosystem"
python3 rhetoric/scripts/rhetoric.py plan "Microservices improve reliability" --no-bridge

Requires: INCEPTION_API_KEY environment variable (Mercury API)

Options:

  • --no-bridge: Skip analogy bridge generation
  • --contradict <evidence>...: Known contradicting evidence for cross-reference validation

validate <file>

Validate a pre-built Toulmin argument graph from a JSON file. No API needed.

python3 rhetoric/scripts/rhetoric.py validate path/to/graph.json

demo

Run built-in demonstration cases showcasing the validation engine. No API needed.

python3 rhetoric/scripts/rhetoric.py demo

Demonstrates detection of: valid induction, hasty generalization, formal fallacies (affirming the consequent), cherry-picking (cross-reference integrity), and surface analogies.

Validation Passes

The engine runs four deterministic validation passes (no LLM calls):

Pass Name Detects
1 Structural completeness Missing claim/data/warrant, absent backing/qualifier/rebuttals
2 Inferential type validation Formal fallacies, hasty generalization, false cause, surface analogy
3 Cross-reference integrity Unaddressed contradictions, unused supporting evidence
4 Qualifier calibration Overclaim/underclaim relative to evidence strength

Required Credentials

INCEPTION_API_KEY=...   # Mercury API — only needed for 'plan' command

The validate and demo commands require no API keys.

Constitution

See constitution.md for 6 inviolable rules governing the argumentation engine.

Synergies

  • -> Encyclopedia: Provide contradicting/supporting evidence for validation Pass 3
  • -> Inland Empire: Store significant argument analyses as memories
  • <- Volition: Complex actions can request argument validation first