Scoville Scribe Anti-AI-Slop

Better wording is useful. Better wording that quietly changes the facts is not.

It usually looks harmless:

  • "May reduce latency" becomes "will improve performance." Smoother, stronger, and no longer the same claim.
  • A product setting named Padding becomes Spacing because repetition felt inelegant. The interface remains stubbornly literal.
  • A summary keeps the result but drops "only when the cache is warm," which was the part keeping the result true.
  • An error tells the user to retry even though the product offers no retry. Encouragement is not yet a control.

That is writing slop: the prose gets cleaner while meaning, terminology, or behavior moves underneath it. The sentence is delighted. The product less so.

Scoville Scribe is an Agent Skill for requested text artifacts and transformations: drafting, editing, rewriting, summarizing supplied sources, localizing, source-exact work, interface text, and wording audits. It preserves meaning, evidence, attribution, canonical terms, schemas, and behavior while removing filler or fixing the requested wording problem. Ordinary answers, explanations, status updates, and domain results stay with the host or their domain owner; merely containing prose is not an activation signal.

Why "Scoville"?

The family is named for useful signal that survives dilution. In writing, the heat is the original meaning after smoothing, shortening, localization, and the occasional thesaurus have all had access to the sentence.

How to use

Name Scoville Scribe when wording must improve without changing its factual or technical contract:

Use Scoville Scribe to tighten this release note while preserving every claim, version number, condition, and uncertainty. Keep the existing product terminology.

Use Scoville Scribe to localize these interface strings into German. Preserve placeholders, access keys, ICU branches, and the distinction between labels, help text, and errors.

Use Scoville Scribe in Source-exact mode to extract the text between the named markers. Preserve the selected bytes and return nothing else.

Explicit $scoville-scribe-anti-ai-slop invocation also works on hosts that support named Skill invocation.

Install

Use an Agent Skills-compatible host and Terra 5.6 Medium or a comparably capable executor such as Opus 4.8. Ask the agent to install:

Install this Agent Skill and refresh the available Skill list:
https://github.com/benjaminstelzer/scoville-scribe-anti-ai-slop/tree/main/scoville-scribe-anti-ai-slop
Keep the installed directory name scoville-scribe-anti-ai-slop. Use Terra 5.6 Medium or a comparably capable executor such as Opus 4.8.

The final path must end in <skills-dir>/scoville-scribe-anti-ai-slop/SKILL.md. For Claude Code, use ~/.claude/skills/ globally or .claude/skills/ inside one project. Other hosts use their supported Skills directory.

What it costs. Activating Scribe loads its Core and then only the Interface, Prose, or Fidelity guidance required by the artifact. That context buys factual fidelity, terminology control, and source-exact handling. Ordinary conversation does not activate Scribe; use the Skill when wording itself is the deliverable or transformation target. Historical token measurements remain bound to their qualified package versions. See benchmark evidence. The family run ledger shows the complete count.

What it enforces

  • Facts survive the edit. Numbers, quotations, conditions, attribution, modality, and uncertainty keep their meaning.
  • Canonical terms stay canonical. A setting named Padding does not become Spacing because the thesaurus was feeling helpful.
  • Working strings keep working. Placeholders, ICU branches, access keys, shortcuts, schemas, and accessible names retain their contracts.
  • Behavior-bound text stays true. Interface labels, help, errors, and procedures describe supported behavior rather than desired fiction.
  • The author's position survives. Voice may improve without inventing certainty, experience, identity, or conclusions.
  • The requested operation stays narrow. An audit reports. An edit changes the smallest real defect. Source-exact output remains exact.

The complete contract is in SKILL.md.

How it works

The Skill description first keeps ordinary conversation and domain-owned normal results out of Scribe. Once an artifact or transformation activates the Core, it resolves truth, terminology, audience, and requested transformation per segment, then loads only the Interface, Prose, or Fidelity guide needed for that segment. Chat delivery alone neither activates nor suppresses Scribe.

Scoville family

Each Skill works independently. Combine only the concerns the task actually needs:

  • Brainstorm explores materially different mechanisms before selection.
  • Research turns web, GitHub, and scholarly evidence into a decision-ready, claim-traceable result.
  • Code owns engineering scope, implementation, risk, and validation.
  • UI owns interface hierarchy, framework fit, accessibility, and rendered evidence.
  • Scribe owns wording, terminology, factual meaning, and source fidelity.
  • Plan owns durable Plans, Work Items, Decisions, and lifecycle state.
  • Handoff transfers active work to another agent or session.

Status

Historical qualification used a reliability-first extension of Microsoft SkillOpt across 1,201 optimization and evaluation runs for the six Scoville Skills. The previous Scribe package passed 30/30 final cases and reduced its loaded Core by 19.84% from v1.0.6. Those results remain evidence for the exact historical package, not for the changed activation contract. See benchmark evidence.

The changed activation contract passed its focused static and composed routing gates. The evaluation does not claim a measured token or latency reduction; the corrected ordinary comparison loaded no Skill in either condition. See the task-scoped routing qualification.

Sources

License

MIT - see LICENSE.