argument-deconstruct

name: argument-deconstruct description: | Reverse-engineer a long-form argument (op-ed, proposal, manifesto, paper) — claim-grounds-warrant chain, hidden warrants, missing rebuttals, argument map. Use for '拆解這個論證' or 'find the warrant'. Hidden assumptions → assumption-surface.

Argument Deconstruct

Reverse-engineer the logical structure of an argument. Where artifact-deconstruct covers all design layers across 6 lenses, this skill goes deeper into one layer: the argument itself.

The critical move: make the hidden warrant explicit. Most arguments hide their warrant; surfacing it is the deconstruction.

When to use

Trigger phrases (any language):

• 「拆解這個論證」「這份提案論證哪裡弱」「找隱性 warrant」
• "deconstruct this argument", "find the warrant", "where does this argument fail"
• "is this argument valid" / "what's the hidden assumption in this claim"

Skip when:

• Artifact has no argumentative structure (descriptive / narrative / reference) — use artifact-deconstruct
• User wants hidden-assumption hunting across many claims — use assumption-surface
• Target is < 200 words — there is not enough argument to deconstruct
• Target is code — use sourceatlas

Workflow

Five steps:

Argument Deconstruction Progress:
- [ ] Step 1: Identify the central claim (and any sub-claims)
- [ ] Step 2: Run Toulmin model on each claim
- [ ] Step 3: Run Burke pentad on the argument as narrative
- [ ] Step 4: Generate argument map (mermaid)
- [ ] Step 5: Self-check — warrant surfaced? rebuttal accounted for?

Step 1: Identify the central claim

Read the artifact and answer:

• What is the main claim? State it as a single sentence beginning with "The text claims that…"
• What sub-claims support it? List them as bullets.
• What is the conclusion the writer wants the reader to reach? This may differ from the explicit claim — the writer may be asking the reader to act, not just believe.

If you cannot extract a clear central claim, the artifact may not be argumentative. Stop and route to artifact-deconstruct instead.

Step 2: Run Toulmin on each claim

For the central claim and each sub-claim, fill in the 6-component model. Read references/lens-toulmin.md for the full method.

Quick form:

Critical rule: If your warrant entry just restates the grounds, you have not surfaced the warrant. The warrant is the bridge — usually a generalization the writer takes for granted.

Step 3: Run Burke pentad

For the argument viewed as narrative, identify five elements. Read references/lens-burke-pentad.md for the full method.

Quick form:

• Act / Scene / Agent / Agency / Purpose
• Claimed ratio: which two dominate by surface emphasis?
• Actual ratio: which two dominate by deeper structure?
• Discrepancy: if claimed ≠ actual, the argument is doing motive-laundering

Step 4: Generate argument map

Use protocols/argument-mapping.md for the mermaid format. The map shows:

• Central claim at the top
• Sub-claims branching down
• Grounds attached to each claim
• Warrants explicitly labeled in dotted lines (since they are usually hidden)
• Missing rebuttals marked with ⚠️

Step 5: Self-check

Before delivering:

• Warrant surfaced: did I write each warrant as a sentence starting with "Because…", and would a reasonable opponent contest it?
• Rebuttal accounted for: did I note whether the argument acknowledges counter-arguments, and if not, what counter-arguments would matter?
• Qualifier present: did I note whether the claim has conditions, or whether it overreaches with "always / everyone / never"?
• Pentad ratio identified: did I name the dominant two pentad elements, and did I note any discrepancy between claimed and actual?
• Argument map renders: does the mermaid block produce a readable graph?
• Ethical position assigned (if applicable): if the argument uses persuasion mechanisms beyond logical argumentation, did I name them and assign 🟢/🟡/🔴/⚫?

Lenses available

This skill ships with 2 self-contained lens references:

• references/lens-toulmin.md — Toulmin argument model (1958)
• references/lens-burke-pentad.md — Burke dramatistic pentad (1945)

Both are also present in artifact-deconstruct/references/lens-rhetoric.md (which combines them). The duplication is intentional per ADR-0002 — each skill is independently loadable.

Output format

# Argument Deconstruction: <text title>

## Central claim
<one sentence>

## Toulmin layout

| Component | Content |
|---|---|
| Claim | ... |
| Grounds | ... |
| **Hidden warrant** | Because ... |
| Backing | ... |
| Rebuttal acknowledged? | yes/no — describe |
| Qualifier present? | yes/no — describe |

## Sub-claims (if any)
<repeat Toulmin per sub-claim>

## Burke pentad
- Act: ...
- Scene: ...
- Agent: ...
- Agency: ...
- Purpose: ...
- **Claimed ratio**: <pair>
- **Actual ratio**: <pair> (note any discrepancy)

## Argument map

\```mermaid
flowchart TD
    Claim[Main Claim] --> Sub1[Sub-claim 1]
    Sub1 -.warrant.-> W1[Hidden warrant: ...]
    Sub1 --> G1[Grounds: ...]
    Claim --> Sub2[...]
    ...
\```

## Findings
- Strongest move: ...
- Weakest move: ...
- Hidden warrant most worth contesting: ...
- Missing rebuttal that matters most: ...

## Bottom line
<One sentence: argument is **strong / sound-but-narrow / hand-waving /
manipulative** because **X**.>

Sample fixtures

• assets/sample-op-ed-AI-regulation.md
• assets/sample-vc-pitch-memo.md

Anti-patterns

• Restating grounds as warrant: a warrant must be a bridge generalization, not a restatement of the data
• Inventing rebuttals where none belong: if the argument acknowledges no rebuttal, say so; do not pretend it does
• Pentad without ratio: just listing 5 elements is bookkeeping; the ratio reveals motive
• Argument map without warrants labeled: the map's value is showing the hidden warrants — if you draw only claim→grounds, you reproduced the surface
• Missing ethical position when persuasion exists: if the argument also uses Cialdini-style appeals, assign 🟢/🟡/🔴/⚫