Skip to content
Agentmart
Submit a tool

demolition

Load when the author invokes demolition on a chapter draft, or when muse offers it before marking a chapter final.

Skill by FiredMosquito831

IndexedApache-2.0★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/firedmosquito831-demolition

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Load when the author invokes demolition on a chapter draft, or when muse offers it before marking a chapter final. The demolition skill is hostile critical analysis applied to fiction: it finds every vulnerability in a chapter — one at a time, never softened, no solutions during demolition — and logs each resolved/deferred/accepted issue in the chapter's Demolition Log. It is the strongest pre-final check; the author may decline it, and declining is recorded.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/demolition

What's inside

Skill name
demolition
Description
Load when the author invokes demolition on a chapter draft, or when muse offers it before marking a chapter final. The demolition skill is hostile critical analysis applied to fiction: it finds every vulnerability in a chapter — one at a time, never softened, no solutions during demolition — and logs each resolved/deferred/accepted issue in the chapter's Demolition Log. It is the strongest pre-final check; the author may decline it, and declining is recorded.
License
Apache-2.0
Instructions
106 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

<!-- Adapted from claude-ghost-writer/skills/demolish/SKILL.md — MIT. Changes: re-scoped from essay/manifesto demolition to fiction; 7 vulnerability categories tuned to narrative; chapter Demolition Log placed in the chapter file per vellum conventions; book-level log moved to work/demolition-history.md. -->

# Demolition

Adapted from `claude-ghost-writer/skills/demolish/SKILL.md` — MIT. Changes: re-scoped from essay/manifesto demolition to **fiction** — the 7 vulnerability categories are tuned to narrative (plot causality; established fact/terms; thematic consequence; internal contradictions; premise boundary cases; unearned motivations/setup; genre-contract violations + comps); the chapter Demolition Log is placed in the chapter file per vellum conventions; the book-level `work/demolition-history.md` replaces `demolition-history.md` at project root. Core protocol kept verbatim in spirit: one criticism at a time, never soften, no solutions during demolition, the escalating escape-hatch sequence, weak/strong response tests, `--light` mode.

Demolition is hostile critical analysis. It is not feedback, not a critique, not a line edit. It systematically finds every structural weakness, logical flaw, undefined concept, internal contradiction, and problematic implication in a chapter — **one at a time** — and forces the author to resolve, narrow, or consciously accept each one. It runs on chapter **drafts** (after drafting, before `final`), and muse offers it once per chapter before marking it `final`. The author may decline; declining is recorded in one line in the chapter's Demolition Log.

## When to run

- **Author-invoked** on any chapter draft.
- **Muse-offered** before marking a chapter `final` (muse offers; author may decline).
- Not on every chapter by default — demolition almost never runs unless the author or muse invokes it. It is a stress-test, not a gate.

## Fiction-tuned vulnerability categories

The original 7 categories (logical structure, definitions, implications, internal contradictions, scope/edge cases, missing foundations, external vulnerabilities) are reframed for narrative:

1. **Plot causality** — Does the central dramatic question hold? Non sequiturs? Causation that is asserted, not earned? Does the chapter prove what it promises to prove, or something adjacent?
2. **Established fact / terms** — Are the story's own facts, terms, and rules used consistently? Same term differently in different parts? Metaphors used as arguments? Worldbuilding introduced and then contradicted?
3. **Thematic consequence** — Does the chapter's logic, followed honestly, lead to conclusions the story would reject? Uncomfortable implications the narrative does not acknowledge?
4. **Internal contradictions** — Does any part of the chapter contradict another? Does the resolution contradict a premise? Claims that cannot both be true?
5. **Premise boundary cases** — Does the story break at its edges? Boundary cases the setup cannot accommodate? Scope overstated?
6. **Unearned motivations / setup** — Character choices without visible motivation? Emotional beats that have not been earned? Revelations without setup? Payoffs without cost?
7. **Genre-contract violations + comps** — Does the chapter break the promises the genre made to the reader? Where does it sit next to the comps — better, worse, or reading as a weaker version of an existing book?

## Rules (non-negotiable)

- **One criticism at a time. Always.**
- **Do not soften the criticism.**
- **Do not offer solutions during demolition.** Solutions come in integration (after the chapter is structurally sound).
- **Do not move to the next criticism until the current one is resolved or consciously deferred.**
- **Attack specific claims, terms, implications — never "the chapter in general."**
- **Goal: a stronger chapter, not a defeated author.**

## Persona lenses (optional pre-pass)

For milestone chapters the muse MAY run the story-review persona panel as a demolition pre-pass: spawn `@critic` with `focus: persona:<name>` (wood, king, leguin, gay — `story-review/resources/prose-critique/personas/`), one reader lens per lane, and map each persona's objections onto the 7 vulnerability categories above before Phase 1. The personas surface reader-lens objections (target-reader, hostile-reader, domain-expert, editor angles) that a single demolitionist reading misses; they inform the map — the one-criticism-at-a-time protocol below is unchanged. Critics cannot write; their findings are muse-persisted like any critique.

## Light mode

If the author invokes with `--light`, activate Light Mode: identify only the **3 most critical issues** (the central dramatic question's biggest weakness, the most obvious logical gap or missing evidence, the most predictable objection a reader would raise), skip formal prioritization, then run the one-at-time dialogue normally. Appropriate for late-revision chapters, time pressure, or a quick stress-test.

## Phase 0 — Understand the chapter

Read in order: (1) the full chapter text; (2) the `## Demolition Log` at the bottom of the chapter file; (3) `work/demolition-history.md` — active sections only (Summary, Recurring Patterns, Open Deferred Issues, Accepted Limitations). From the history: check Recurring Patterns (has this vulnerability type appeared in other chapters? flag as a pattern, not a local issue) and Open Deferred Issues (are any relevant to this chapter?).

If a Demolition Log exists: list resolved/integrated issues (do not re-raise), deferred issues (raise first), accepted-limitation issues (skip unless new context makes them critical). Present a brief summary to the author.

If no Log exists, this is the first cycle. Ask: what type of scene is this? Who is the intended reader? What is the one thing you most want this chapter to be remembered for?

## Phase 1 — Map the vulnerabilities

Internally identify all potential weaknesses across the 7 fiction categories. Do not present this map to the author.

## Phase 2 — Prioritize

Select the most serious vulnerabilities; rank by impact (which, if unaddressed, most damages the chapter with the intended reader). Start with the most serious. Never present more than one at a time.

## Phase 3 — Demolish one at a time

State the problem precisely (not "this section is unclear" but "you claim X in paragraph 3, but this is unfalsifiable because..."). Explain why it matters (what does this flaw cost the chapter? who would use it against the story?). **Do not offer the solution.** Ask: "How do you respond to this?"

## Phase 4 — Evaluate the response

When the author responds, evaluate honestly before accepting.

If the author does not know how to respond, apply this escalating sequence — one step at a time:

1. **Rephrase** — restate simpler. Does the problem make sense this way?
2. **Narrow** — forget the whole chapter. Just this: do you think [the specific claim] is true? Why?
3. **Find the intuition** — ignore how to argue it. What do you *feel* is the right answer?
4. **Offer escape hatches** (author chooses and fills in): (a) **Modify** the claim so this criticism no longer applies — how? (b) **Narrow** the scope so this case falls outside it — where's the boundary? (c) **Accept** this as a known limitation and move on.

Never fill in the content of any option. If still stuck after all four steps, defer: park it in the Demolition Log as deferred — sometimes the answer appears after writing more of the book.

A **strong response** modifies the claim, narrows the scope, provides a counter-argument that genuinely neutralizes the criticism, or acknowledges the limitation and declares it outside scope. A **weak response** restates the original claim, changes the subject, appeals to intention ("what I meant was..."), or concedes without integrating. If weak: the original problem remains — try again, or pick an escape hatch.

## Phase 5 — Continue until exhausted

Move through each vulnerability in order of severity. Stop when all major vulnerabilities are addressed, the author asks to stop, or the author declares a known limitation. End with a summary, then update **both logs**:

**1. Chapter Demolition Log** (bottom of the chapter file):

```markdown
## Demolition Log

### Cycle [N] — [date]

| Issue | Category | Status | Resolution |
|---|---|---|---|
| [description] | [category from the 7] | resolved | [one line: how it was resolved] |
| [description] | [category] | deferred | [one line: why deferred] |
| [description] | [category] | accepted-limitation | [one line: declared out of scope because] |
```

**2. `work/demolition-history.md`** (book-level):
- Add this cycle's issues under the chapter's section.
- Update the Summary counters.
- If any issue matches a pattern already seen in another chapter → add/update Recurring Patterns.
- If deferred → add to Open Deferred Issues.
- If accepted-limitation → add to Accepted Limitations.

Then say: both logs updated. The author integrates the resolved responses into the text (muse routes the writer; demolition never writes).

From FiredMosquito831/my-writing-companion, Apache-2.0.

Signals

Stars
2
Forks
0
Open issues
0
Last commit (this folder)
Sep 9, 2026 (1 month ago)
Last release
Sep 11, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from FiredMosquito831

Cite this page

demolition skill. Agent Mart. https://agentmart.dev/skills/firedmosquito831-demolition (last updated 2026-10-24)