| name | deslop-prose |
| description | Clean academic or professional prose that carries AI-generated residue by removing hype, decorative structure, vague metadiscourse, pseudo-technical phrasing, and claim-evidence mismatch while preserving technical meaning, qualifications, and authorial intent. |
Deslop Prose
Edit final or near-final prose so it says the same thing with higher signal and lower inflation.
Use this when the draft is already directionally correct but reads as generic, over-polished, over-enumerated, or AI-shaped. Do not use it for factual verification, source-of-truth adjudication, or discussion-history cleanup.
Use when
Use this skill when:
- academic or professional prose sounds inflated, generic, or mechanically polished
- claims are broader or stronger than the evidence stated nearby
- decorative structure or metadiscourse is displacing content
- terminology sounds improvised when an established term would be clearer
Do not use
Do not use this skill for:
- issue, PR, or prior-draft residue cleanup where the main problem is discussion provenance
- source-of-truth auditing, factual verification, or citation checking
- legal or policy text where rhetorical softening could change meaning
- drafts that still need architectural or substantive decisions
- quantifier scope, theorem/lemma/proposition hierarchy, or claim-level mathematical relations — this is logical content, not rhetorical inflation; use
math-claim-integrity
- Japanese-language phrasing anti-patterns — use
wabun-math-style
- judging whether a research direction or connection is worth pursuing — use
research-significance
Use deslop-history when the artifact leaks process history, superseded alternatives, or agent-facing framing.
Inputs
Gather or infer these before editing:
artifact — the prose to clean
intended_reader — who must read or act on it
artifact_type — paper, memo, report, proposal, README, summary, or similar
technical_terms — terms that should remain stable
evidence_scope — the facts, results, citations, or qualifications that bound the claims
Ask only when rewriting could blur a technical term, scope boundary, or required qualification.
Procedure
- Identify the intended reader, artifact type, and evidence scope.
- Mark sentences where style is carrying more weight than content.
- Tighten claims so their strength matches the stated support.
- Remove decorative structure, generic metadiscourse, and pseudo-technical inflation.
- Rewrite with stable terminology and direct claims.
- Check that qualifications, citations, definitions, and authorial intent still hold.
Target Patterns
hype-inflation — unsupported importance claims such as "transformative" or "pivotal"
enumeration-inflation — decorative lists that imply coverage without support
generic-metadiscourse — scaffolding like "It is important to note that"
framework-inflation — grand labels such as "broader landscape" or "holistic framework" without a concrete framework
methodological-theatre — vague rigor language where a concrete method should be named
vague-significance — "valuable insights" or "important implications" without the concrete insight or implication
defensive-negative-framing — "This is not X" when a positive statement would be clearer
unstable-terminology — unusual paraphrases that obscure standard concepts
hyphen-chain-pseudo-terms — long ad hoc compounds that sound technical without adding precision
formulaic-symmetry — repeated structures like "not only X but also Y" when the contrast adds no value
claim-evidence-mismatch — certainty, breadth, or importance exceeding the support in the text
Rewrite Rules
- Prefer concrete contribution claims over broad importance claims.
- Prefer one supported dimension over a long decorative list.
- Start with the claim, not the sentence-level announcement of the claim.
- Name the actual method, evidence, dataset, theorem, metric, proof step, or procedure when known.
- Preserve necessary hedging, qualifications, citations, notation, and scope boundaries.
- Preserve established technical terms; do not paraphrase for freshness.
- Rewrite long ad hoc hyphenated compounds as ordinary noun phrases unless the compound is standard.
- Keep negative framing only when it prevents a likely misuse.
- Do not turn precise prose into smooth but lower-density prose.
- Do not change quantifier scope, hypothesis strength, or theorem/lemma/proposition naming and hierarchy while tightening a sentence — that is a logical-content edit, not a style edit.
Minimal examples:
Before: This transformative framework provides valuable insights into a broad landscape of practical applications.
After: This framework identifies the practical conditions under which the method applies.
Before: We systematically and rigorously analyse the method across efficiency, scalability, robustness, interpretability, and applicability.
After: We evaluate runtime and robustness on the reported benchmarks.
Must not flag: The estimator is asymptotically efficient and consistent under standard regularity conditions.
Unchanged — "asymptotically efficient," "consistent," and "regularity conditions" are established statistical terms, not inflation, despite the sentence's density.
Owned by a neighbouring skill, not this one: We prove three new theorems, but one is used only to establish the other two.
This is a theorem-hierarchy defect (independent results conflated with proof infrastructure), not prose inflation — use `math-claim-integrity` (rule R-F), not `deslop-prose`.
Output
If the user asks to edit a file, apply the rewrite directly. If the user asks for cleanup text only, return the revised prose without an audit log.
When the user asks for review before editing, provide:
- the inferred reader and artifact type
- the main inflation patterns present
- the cleaned prose
When editing repository files:
- modify only the target artifact or directly required index entries
- keep the diff focused on prose cleanup guidance
- do not introduce style taxonomies beyond what the artifact needs
Quality Check
Before finishing, verify:
- the claim strength matches the evidence actually present
- decorative enumeration and generic metadiscourse are gone
- technical meaning, qualifications, and terminology remain intact
- the prose sounds more direct, not more impressive
deslop-history would not be the more appropriate skill
- no quantifier, hypothesis, or theorem-hierarchy content changed in the course of tightening a sentence
Relationship to Other Skills
- Use
deslop-history when the main problem is discussion provenance, prior-draft residue, or process framing.
- Use
sot-integrity when authority, factual grounding, or trust scope must be audited.
- Use
textlint-cli for deterministic lint checks, not context-sensitive prose judgment.
- Use
math-claim-integrity for quantifier scope, theorem/proposition hierarchy, proof/computation honesty, or other logical-structural defects; this skill never edits claim-level mathematical content.
- Use
wabun-math-style for Japanese-language anti-patterns (particle misuse, unnatural literal translation, epistemic hedges); this skill is language-agnostic and does not target Japanese-specific phrasing.
- Use
research-significance to assess whether a research direction or connection has scholarly value; this skill only cleans prose after that judgment is made.