tylergarlick/abraxas/logos

Logos

Logos is argument anatomy — the systematic mapping of what an argument actually contains before it is evaluated, debated, or accepted. It solves the problem of invisible structural weakness: LLMs generate confident conclusions from arguments with missing premises, skipped inferences, hidden assumptions, and unfalsifiable claims.

Debug & Reasoning Standard: This skill adheres to the references/reasoning-standard.md protocol. All mapping and gap analysis must be preceded by a <thought> block detailing the structural hypothesis. Each identified gap or assumption must be explicitly anchored to the argument text using the [Reasoning: {TextFragment}] -> {GapType} pattern.

Logos makes argument structure visible. It is the mandatory pre-analysis layer that feeds into Janus (for epistemic labeling) and Agon (for adversarial testing).

The Core Problem Logos Solves

Large Language Models generate conclusions from structurally flawed arguments with perfect confidence. An argument may contain:

• Unstated premises taken as given without acknowledgment
• Inference steps that skip logical steps (enthymemes)
• Hidden assumptions that would not survive scrutiny
• Missing evidence that would change the conclusion
• Circular reasoning that appears linear
• Unfalsifiable claims that cannot be tested

Without structure, there is no surface for epistemic labeling (Janus) or adversarial testing (Agon) to grip. Logos provides that surface.

When to Use Logos

Use Logos when:

• You need to understand what an argument actually contains before evaluating it
• You want to find the hidden assumptions in a claim
• You are preparing for an Agon debate and need to map the argument first
• You want to test whether a claim is even testable (falsifiability)
• You need to identify the weakest point in an argument's structure

Use Logos before Agon:

1. User has claim or argument
         │
         ▼
2. /logos (alias: /abraxas:logos, /abx:logos)   map        → Extract structure
3. /logos (alias: /abraxas:logos, /abx:logos)   gaps       → Identify weaknesses  
4. /logos (alias: /abraxas:logos, /abx:logos)   assume     → Surface assumptions
5. /logos (alias: /abraxas:logos, /abx:logos)   falsify    → Test validity
6. /logos (alias: /abraxas:logos, /abx:logos)   report     → Get recommendation
         │
         ▼ (if ready for debate)
7. /agon (alias: /abraxas:agon, /abx:agon)   debate {refined claim}

Do not use Logos when:

• The input is purely factual and already well-supported — Logos is for analysis of argument structure, not fact-checking
• The input is symbolic, dream material, or [DREAM] — use Abraxas Oneironautics instead
• The input is a question without an argument structure — Logos requires premises and conclusions to map

The Six Commands

/logos map — Map Argument Structure

Extract and organize the structural components of an argument.

Required argument: The argument or claim to map (can be a sentence, paragraph, or multi-paragraph argument)

Output includes:

• All explicit premises (P1, P2, ...)
• Intermediate conclusions (IC1, IC2, ...)
• Final conclusions (C1, C2, ...)
• Inference chain showing how premises connect to conclusions
• Classification of each inference type (deductive, inductive, abductive, probabilistic, analogical)

Example:

/logos (alias: /abraxas:logos, /abx:logos)   map All dogs bark. Max is a dog. Therefore, Max barks.

Output:

[LOGOS — MAP]
 
PREMISES
────────
P1: "All dogs bark" — [explicit]
P2: "Max is a dog" — [explicit]
 
INTERMEDIATE CONCLUSIONS
───────────────────────
(none)
 
FINAL CONCLUSIONS
─────────────────
C1: "Max barks"
 
INFERENCE CHAIN
───────────────
P1 + P2 → C1 (via deductive — modus ponens)
 
[STRUCTURE: VALID]
All premises explicit, inference valid.

/logos gaps — Identify Missing Elements

Analyze the argument map for missing structural elements that would strengthen or weaken the argument.

Optional focus: "premises", "evidence", "assumptions", "inferences", or "all" (default)

Output includes:

• Premise gaps (missing evidence, unstated sub-premises)
• Inference gaps (skipped steps, hidden assumptions)
• Conclusion gaps (overgeneralizations, unqualified claims)
• Evidence gaps (supporting evidence absent, counter-evidence unaddressed)
• Severity rating: High / Medium / Low

Example:

/logos (alias: /abraxas:logos, /abx:logos)   gaps Remote work is better because employees are more productive at home.

Output:

[LOGOS — GAPS]
 
PREMISE GAPS
────────────
P1: "employees are more productive at home"
  Gap: No evidence cited for productivity claim
  Significance: HIGH
  Bridge: Cite productivity studies comparing remote vs office
 
INFERENCE GAPS
──────────────
P1 → C1
  Gap: Assumes productivity directly implies "better"
  Type: hidden assumption — value judgment unstated
  Fix: Define "better" (better for whom? in what metrics?)
 
CONCLUSION GAPS
───────────────
C1: "Remote work is better"
  Gap: No scope limitation — better for all contexts?
  Type: overgeneralization
  Qualification needed: "for roles with X characteristics, Y may apply"
 
[GAP SEVERITY SUMMARY]
High: 2 — Fundamental weaknesses
Medium: 1 — Notable weaknesses
Low: 0

/logos inferences — Trace Inference Chains

Map the complete path from premises to conclusions, showing each step in detail with epistemic assessment.

Optional focus: specific premise or conclusion to trace from

Output includes:

• Each inference step with inputs and outputs
• Inference type (deductive, inductive, etc.)
• Hidden assumptions required for each step
• Validity assessment: Valid / Invalid / Uncertain
• Overall chain validity
• Alternative chains not taken

Example:

/logos (alias: /abraxas:logos, /abx:logos)   inferences Climate change is caused by humans because CO2 levels have increased and CO2 causes warming.

/logos assume — Surface Hidden Assumptions

Explicitly identify and examine the unstated assumptions required for an argument to hold.

Output includes:

• Each hidden assumption identified
• Type: Factual / Definitional / Normative / Methodological
• What would falsify each assumption
• Current evidence status
• Assessment: Plausible / Questionable / Unsupported / Dubious
• Argument dependency (how many assumptions the conclusion depends on)

Example:

/logos (alias: /abraxas:logos, /abx:logos)   assume We should switch to renewable energy because it's the right thing to do.

/logos falsify — Test Argument Validity

Attempt to falsify the argument by finding conditions under which the conclusion would be false.

Output includes:

• Falsification conditions for each premise
• Falsification conditions for each inference
• Falsification conditions for the conclusion
• Evidence status: Checked / Unchecked / Unknown
• Falsifiability assessment: Falsifiable / Conditionally Falsifiable / Not Falsifiable / Underspecified

Example:

/logos (alias: /abraxas:logos, /abx:logos)   falsify This theory is true because the scientist said so.

Output:

[LOGOS — FALSIFY]
 
PREMISE FALSIFIERS
──────────────────
P1: "the scientist said so"
  Falsification condition: Scientist is not an expert in relevant domain
  Evidence status: UNCHECKED
  If false: Entire argument collapses
 
INFERENCE FALSIFIERS
────────────────────
P1 → C1
  Breaking condition: Authority claim without evidence is fallacious
  Test: Check scientist's credentials and whether they cited evidence
 
CONCLUSION FALSIFIERS
─────────────────────
C1: "This theory is true"
  Falsification condition: Any contradictory evidence exists
  Would still require: Show contradictory evidence is valid
 
[FALSIFIABILITY ASSESSMENT]
Claim is: NOT FALSIFIABLE
 
Rationale: Argument relies on authority, not evidence. No testable 
conditions specified. This is an argument-from-authority fallacy.
 
[RECOMMENDATION]
This argument is not suitable for Agon debate. It must be 
restructured with evidence before analysis can proceed.

/logos report — Generate Structured Analysis

Produce a comprehensive report combining all Logos analyses into a single deliverable.

Optional format: "brief", "standard" (default), or "detailed"

Output includes:

• Executive summary (2-3 sentences)
• Structural assessment (premise count, inference count, overall validity)
• Gap severity rating
• Falsifiability assessment
• Assumption profile summary
• Recommendation for Agon (ready / needs refinement / too weak)

This is the command to run before `/agon debate`.

Integration Points

Janus Integration

Logos produces argument maps that feed into Janus for epistemic labeling:

Workflow:

/logos (alias: /abraxas:logos, /abx:logos)   map {argument} + /sol (alias: /abraxas:sol, /abx:sol)   label

Agon Pre-Layer

Logos is the mandatory pre-layer to Agon debates. Always run /logos report before /agon debate.

Edge Cases

Circular Reasoning

If the inference chain loops back to a premise:

[CIRCULAR REASONING DETECTED]
This is a fallacy — the argument assumes what it claims to prove.
Cannot proceed to Agon. Remove circularity first.

Loaded Questions

If the argument is embedded in a question that presupposes its answer:

[LOADED QUESTION DETECTED]
Restate as a neutral assertion before analysis.

False Dichotomy

If only two options are presented when more exist:

[FALSE DICHOTOMY DETECTED]
Additional options exist. The argument's conclusion depends on 
eliminating a false choice.

Ad Hominem

If the argument attacks the person rather than the claim:

[AD HOMINEM DETECTED]
Extract the claim from the attack and analyze that instead.

Argument from Authority

If a premise relies solely on authority without evidence:

[AUTHORITY CLAIM DETECTED]
Authority claims require domain expertise, consensus, or cited evidence.
This premise needs additional support.

Pure Assertion

If the input is a claim with no supporting reasoning:

[PURE ASSERTION — NO ARGUMENT STRUCTURE]
No premises, inferences, or reasoning to map.
Provide the reasoning behind this claim to analyze.

Storage

Logos stores argument maps in ~/.logos/:

~/.logos/
├── arguments/       # Saved argument maps (JSON)
├── sessions/        # Session-specific analyses
├── index.md         # Searchable argument index
└── config.md        # User preferences

Argument maps are JSON with full structural data, enabling:

• Persistence across sessions
• Import/export for collaboration
• Query by Janus and Agon

Constraints

1. Always map before debate: Do not run /agon debate without first running /logos report on the claim.

1. Label honestly: When mapping, identify all weaknesses. Do not minimize gaps to make the argument appear stronger.

1. Flag unfalsifiable claims: If a claim cannot be tested, state that explicitly. Do not treat it as if it could be verified.

1. Surface assumptions: Hidden assumptions are the most valuable output. Prioritize finding them.

1. Use Janus labels: When the user requests labeling, apply [KNOWN], [INFERRED], [UNCERTAIN], [UNKNOWN] appropriately.