Argument Deconstruct

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

Central claim

Toulmin layout

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

Sub-claims (if any)

Burke pentad

• Act: ...
• Scene: ...
• Agent: ...
• Agency: ...
• Purpose: ...
• Claimed ratio:
• Actual ratio: (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

## Sample fixtures

- [`assets/sample-op-ed-AI-regulation.md`](assets/sample-op-ed-AI-regulation.md)
- [`assets/sample-vc-pitch-memo.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 🟢/🟡/🔴/⚫

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [kouko](https://github.com/kouko)
- **Source:** [kouko/monkey-skills](https://github.com/kouko/monkey-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.