Identify the text scope, audience, language, and technical domain from the request and surrounding files.
Read applicable glossaries, style guides, neighboring human-authored text, and exact type or protocol names.
Scan for suspicious terms with known-slop-terms.md, then read every hit in its sentence and domain context.
Keep established terms, directly replace obvious generic filler when edits are authorized, and investigate only consequential or genuinely uncertain terminology.
Re-read each changed sentence for meaning and grammar, then search the edited scope for missed variants.

Decision Test

For each candidate, ask:

Is it an exact product, protocol, legal, academic, or code-defined term?
Does it express a distinction that a simpler replacement would lose?
Is it normal for this audience and genre?
Is it consistent with the project's deliberate vocabulary?
Would removing or replacing it make the sentence more specific?

Classify the result:

keep: established or meaningfully precise in context;
replace: inflated, mistranslated, vague, or borrowed from the wrong domain;
rewrite: the problem is the whole construction, not one word;
uncertain: available evidence does not justify a confident change.

Do not turn this into a blacklist pass. For example, contract, surface, artifact, canonical, and envelope are correct in some technical contexts and misleading in others.

Evidence Proportional to the Decision

Use evidence when a term is domain-specific, disputed, unfamiliar, externally defined, current, or consequential. Follow verification-procedure.md then.

Prefer:

the governing specification, standard, official API, or project glossary;
the target project's established vocabulary, while checking whether it is itself inconsistent;
representative practitioner or academic usage when official sources do not settle natural phrasing.

A generic cliché such as "it is worth noting" usually does not require five web searches. A proposed replacement for a protocol field, regulated term, or specialist concept may require primary evidence and independent corroboration. When browsing, cite the sources that actually support the terminology decision.

Replacement Rules

Use replacement-patterns.md as a candidate menu, not an automatic mapping.

Preserve technical meaning, requirement level, uncertainty, and the author's intended register.
Prefer the term already used by the governing API or community when it fits the same concept.
Rewrite the full sentence when a one-word substitution creates awkward Korean or English.
Do not alter identifiers, schema fields, commands, quoted source text, trademarks, or external API names unless the user explicitly requested such changes.
Preserve intentionally formal language in legal, policy, academic, or standards writing.

Scope and Autonomy

A review request produces findings and suggested replacements without editing.
An explicit "fix", "replace", or "rewrite" request authorizes changes within the supplied files or text; do not ask for a second confirmation.
Do not expand a paragraph review into a repository-wide terminology migration.
Infer the domain from the text and repository when evidence is clear. Ask only when unresolved domain ambiguity would materially change the replacement.

Output

Lead with the corrected text or the highest-impact findings. For a review with several candidates, use a compact table:

Location	Term	Verdict	Replacement	Evidence or reason

Omit unchanged watch-list hits unless explaining a likely false positive matters. Separate evidence-backed terminology facts from stylistic preference.

For file edits, report changed files, notable terms kept or replaced, evidence looked up, and validation performed.

Validation

Search the edited scope for the exact suspect terms and morphological variants that were meant to change.
Re-read full sentences and neighboring paragraphs; confirm agreement, particles, articles, tense, and requirement level.
Confirm links, code spans, quotations, and identifiers were not accidentally altered.
Inspect the diff for unrelated tone or content changes.

A clean grep is not enough if the replacement changed meaning.

Reference Files

File	Read when	Purpose
known-slop-terms.md	Scanning a substantial text	Context-sensitive watch list
replacement-patterns.md	Drafting alternatives	Candidate replacements and domain exceptions
verification-procedure.md	A decision needs external evidence	Risk-scaled source and search procedure

Gotchas

A term cannot prove AI authorship. Report the wording problem, not an unsupported attribution.
Local repetition may reflect a team convention, copied boilerplate, or repeated generated text; it is preference evidence, not automatic truth.
Simplifying every abstract noun or English loanword can erase necessary technical distinctions.
Current vendor terminology and standards can change; browse when freshness matters.
A less common term may be correct for a specialist audience. Frequency alone is not a verdict.

anti-ai-slop-terminology

Anti AI-Slop Terminology

Quick Start