# Artifact Deconstruct

> |

- **Type:** Skill
- **Install:** `agentstack add skill-kouko-monkey-skills-artifact-deconstruct`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [kouko](https://agentstack.voostack.com/s/kouko)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [kouko](https://github.com/kouko)
- **Source:** https://github.com/kouko/monkey-skills/tree/main/deconstruct-toolkit/skills/artifact-deconstruct

## Install

```sh
agentstack add skill-kouko-monkey-skills-artifact-deconstruct
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Artifact Deconstruct

Reverse-engineer the design decisions behind any polished artifact. The
goal is not summary — it is **design archaeology**: surface what the
creator decided, what frameworks they borrowed, and what they
deliberately omitted.

## When to use

Trigger phrases (any language):

- 「拆解這份」「反推這個」「為什麼這份寫得這麼好」「這份是怎麼設計的」
- "deconstruct this", "reverse engineer", "design behind this", "teardown"
- "why does this work" / "what is this artifact's blueprint"

Skip when:

- User wants a **summary** — use plain reading
- Artifact is < 200 words and not a structured argument — there is not enough design to deconstruct
- Artifact is purely informational reference (Wikipedia, dictionary, raw data) — no *design* to recover
- Target is source code — use `sourceatlas` instead
- Target is the user's own thinking — use `philosophers-toolkit`

## Workflow

Six steps, run in order. Track progress with this checklist:

```
Deconstruction Progress:
- [ ] Step 1: Identify artifact type
- [ ] Step 2: Select lens combination
- [ ] Step 3: Run six-dimension analysis
- [ ] Step 4: Apply selected lenses
- [ ] Step 5: Generate deconstruction report
- [ ] Step 6: Self-check against anti-patterns
```

### Step 1: Identify artifact type

Detect which type the input belongs to. Use signals, not just user
labels.

| Type | Signals |
|---|---|
| Marketing copy / advertising | Short, persuasion-focused, has CTA, hero section |
| Document pack / playbook | Multi-file, README present, role-routed sections |
| Single SOP / how-to | Sequential numbered steps, checklist format |
| Presentation deck | Slide structure (1 slide ≈ 1 idea), visual-heavy |
| Onboarding flow | UI screens or sequential steps, progressive disclosure |
| Long-form article | Argumentative or narrative prose, 1000+ words |
| Speech / political text | First-person plural, emotional cadence, call to collective |
| Literature / film | Narrative, character arcs, symbolic imagery |
| UI screen | Visual layout, interactive elements, status indicators |

Record the detected type. If multiple apply, pick the **dominant** one
and note secondary types in the report.

### Step 2: Select lens combination

Read [protocols/lens-selection.md](protocols/lens-selection.md) for
the full decision tree. Default mappings:

| Artifact type | Lens combination |
|---|---|
| Marketing copy | `lens-persuasion` + `lens-rhetoric` |
| Document pack | `lens-genre` + 6-dimension full pass |
| SOP / playbook | `lens-genre` |
| UI / onboarding | `lens-ux` + `lens-persuasion` |
| Speech / political | `lens-rhetoric` + `lens-frame` |
| Literature / film | `lens-semiotic` + `lens-frame` |
| Presentation deck | `lens-rhetoric` + `lens-genre` |

You may add a third lens if the artifact has unusual depth (e.g., a
landing page that also embeds a long argument benefits from
`lens-rhetoric` + `lens-persuasion` + `lens-genre`). Do NOT use all
six — that signals indecision, not thoroughness.

### Step 3: Run six-dimension analysis

Always run these six dimensions, regardless of which lenses you
selected. They are the structural backbone.

1. **Audience routing** — Who reads what, when?
2. **Creation sequence** — Reading order vs probable writing order
3. **Source genealogy** — What existing frameworks were borrowed?
4. **Rhetorical structure** — How does it persuade?
5. **Design patterns** — What recurring techniques appear?
6. **Negative space** — What is deliberately omitted?

See [protocols/six-dimensions.md](protocols/six-dimensions.md) for the
prompts that operationalize each dimension.

### Step 4: Apply selected lenses

Before reading lens references, **resolve the language variant** for
any of the 4 culturally-sensitive lenses (`lens-rhetoric` /
`lens-persuasion` / `lens-genre` / `lens-frame`) per
[`protocols/lens-variant-selection.md`](protocols/lens-variant-selection.md).
The universal-core `lens-{rhetoric,persuasion,genre,frame}.md` files
contain meta-only routing content; the analytical questions live in
the per-language variant (`-anglo` / `-ja` / `-zh`). `lens-semiotic`
and `lens-ux` do not have language variants in v0.2.0.

Read each selected lens reference (variant-resolved) and apply its
analytical questions to the artifact. Each lens reference contains:

- Primary-source citation (locked)
- Specific analytical questions
- Worked example
- "Do not over-apply" pitfalls

Available lenses (* = language-variant routing required):

- [`references/lens-semiotic.md`](references/lens-semiotic.md) — Barthes 5 codes (HER / PRO / SEM / SYM / REF)
- [`references/lens-rhetoric.md`](references/lens-rhetoric.md) ✱ — universal core; routes to `-anglo` (Burke + Toulmin) / `-ja` (Hinds + kishōtenketsu) / `-zh` (文心雕龍 六觀)
- [`references/lens-frame.md`](references/lens-frame.md) ✱ — universal core; routes to `-anglo` (Goffman + Lakoff) / `-ja` (建前/本音 + ma) / `-zh` (面子/關係 + 陰陽)
- [`references/lens-genre.md`](references/lens-genre.md) ✱ — universal core; routes to `-anglo` (Swales + Bhatia) / `-ja` (序論-本論-結論 + 拝啓 letter) / `-zh` (緒論-本論-結論 + 公文)
- [`references/lens-ux.md`](references/lens-ux.md) — Nielsen-Norman 10 heuristics + Norman affordance
- [`references/lens-persuasion.md`](references/lens-persuasion.md) ✱ — universal core; routes to `-anglo` (Cialdini + Brignull) / `-ja` (Cialdini + JP cultural) / `-zh` (Cialdini + 面子/關係)

**Always state which variant was applied in the report** (e.g., "Applied lens-rhetoric-ja") — see `protocols/lens-variant-selection.md` §"Step 5".

### Step 5: Generate deconstruction report

Use [`assets/report-template.md`](assets/report-template.md). Six
sections:

1. **Surface observations** (what you see)
2. **Design decisions** (audience / sequence / rhetoric)
3. **Borrowed frameworks** (genealogy)
4. **Rhetorical mechanisms** (with ethical position 🟢/🟡/🔴/⚫)
5. **Replicable lessons** (5 concrete takeaways)
6. **Weaknesses or warnings** (missing moves / suspicious warrants / dark-pattern risk)

End with a one-line **bottom-line verdict**.

### Step 6: Self-check

Before delivering, run [`checklists/anti-patterns.md`](checklists/anti-patterns.md)
self-check. Most common failures:

- Output drifted toward summary ("this document covers A, B, C") — must be design-decision focused
- Persuasion mechanisms named but no ethical position assigned
- All six dimensions not covered
- Negative space ignored — only analyzed what IS written
- Single lens tunnel vision — applied one lens to a multi-lens artifact

## Lens Library Quick Reference

All lens files are in `references/`, one level deep. Lenses marked ✱
are language-variant-routed — see `protocols/lens-variant-selection.md`.

| Lens | Variants | Authors (per variant) | When to apply |
|---|---|---|---|
| `lens-semiotic` | (no variants) | Roland Barthes (S/Z, 1970) | Literature, film, advertising imagery, texts with strong subtext |
| `lens-rhetoric` ✱ | -anglo / -ja / -zh | Burke 1945 + Toulmin 1958 / Hinds 1983 + 1987 + Oh 2025 (kishōtenketsu) / 劉勰《文心雕龍》六觀 | Speeches, proposals, political texts, arguments needing structural validation |
| `lens-frame` ✱ | -anglo / -ja / -zh | Goffman 1974 + Lakoff 1980 / + Doi 1971 + Yamamoto 1977 + Markus & Kitayama 1991 (建前/本音, 空気, ma) / + Hu 1944 + Hwang 1987 + Peng & Nisbett 1999 (面子/臉, 陰陽) | Texts with strong metaphors, suspected hidden assumptions, ideological texts |
| `lens-genre` ✱ | -anglo / -ja / -zh | Swales 1990 + Bhatia 1993 / + 木下 1981 + Hinds 1987 (序論-本論-結論, 拝啓 letter) / + 行政院公文程式條例 (緒論-本論-結論, 公文) | Document packs, SOPs, academic papers, business proposals — anything genre-recognizable |
| `lens-ux` | (no variants) | Jakob Nielsen + Don Norman (1988, 2013) | UI screens, app flows, websites, onboarding experiences (yes, docs are UX too) |
| `lens-persuasion` ✱ | -anglo / -ja / -zh | Cialdini 2021 + Brignull 2024 / + Doi 1971 + cross-cultural empirical work / + Hwang 1987 (面子/關係/人情) | Marketing copy, sales pages, landing pages, onboarding, political/policy texts |

## Output template

See [`assets/report-template.md`](assets/report-template.md) for the
full 6-section format.

## Anti-patterns

See [`checklists/anti-patterns.md`](checklists/anti-patterns.md) for
the self-check list. Run before every delivery.

## Sample artifacts (eval fixtures)

- [`assets/sample-dropbox-landing-2024.md`](assets/sample-dropbox-landing-2024.md)
- [`assets/sample-notion-onboarding-pack.md`](assets/sample-notion-onboarding-pack.md)
- [`assets/sample-stripe-signup-flow.md`](assets/sample-stripe-signup-flow.md)

## 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.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-kouko-monkey-skills-artifact-deconstruct
- Seller: https://agentstack.voostack.com/s/kouko
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.