demolition
Skill2 starsChanged 25 days agoApache-2.0View on GitHub
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.
In this repository33 other files it gives its agents.
- character-simSkill
- cold-readSkill
- creative-researchSkill
- creative-writing-craftSkill
- creative-writing-modesSkill
- creative-writing-museSkill
- exportSkill
- gatesSkill
- grill-with-docsSkill
- information-hierarchySkill
- intent-modelingSkill
- kb-integritySkill
- kb-managementSkill
- knowledge-layersSkill
- llm-writingSkill
- md-validationSkill
- project-setupSkill
- qi-layerSkill
- reader-simSkill
- reflectSkill
- series-bibleSkill
- seriesSkill
- shared-daoSkill
- story-ledgersSkill
- story-memorySkill
- story-planningSkill
- story-reviewSkill
- structured-artifactSkill
- style-guardrailsSkill
- voiceSkill
- writing-principlesSkill
- writing-staffingSkill
- zoom-outSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Muse-offered: before marking a chapter final (muse offers.
- Distinctive: line 19
Not on every chapter by default — demolition almost never runs unless the author or muse invokes it.
- Distinctive: line 54
If a Demolition Log exists: list resolved/integrated issues (do not re-raise), deferred issues (raise first), accepted-limitation issues …
- Distinctive: line 56
If no Log exists, this is the first cycle.
- Distinctive: line 72
When the author responds, evaluate honestly before accepting.
- Distinctive: line 74
If the author does not know how to respond, apply this escalating sequence — one step at a time:
- Distinctive: line 81
Never fill in the content of any option.
- Distinctive: line 104
Add this cycle's issues under the chapter's section.
- Distinctive: line 105
Update the Summary counters.
- Distinctive: line 106
If any issue matches a pattern already seen in another chapter → add/update Recurring Patterns.
- Distinctive: line 107
If deferred → add to Open Deferred Issues.
- Distinctive: line 108
If accepted-limitation → add to Accepted Limitations.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<!-- 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:
- 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?
- 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?
- Thematic consequence — Does the chapter's logic, followed honestly, lead to conclusions the story would reject? Uncomfortable implications the narrative does not acknowledge?
- Internal contradictions — Does any part of the chapter contradict another? Does the resolution contradict a premise? Claims that cannot both be true?
- Premise boundary cases — Does the story break at its edges? Boundary cases the setup cannot accommodate? Scope overstated?
- Unearned motivations / setup — Character choices without visible motivation? Emotional beats that have not been earned? Revelations without setup? Payoffs without cost?
- 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:
- Rephrase — restate simpler. Does the problem make sense this way?
- Narrow — forget the whole chapter. Just this: do you think [the specific claim] is true? Why?
- Find the intuition — ignore how to argue it. What do you feel is the right answer?
- 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):
## 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).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

