prompt-repetition

A prompt repetition technique for improving LLM accuracy. Achieves significant performance gains in 67% (47/70) of 70 benchmarks. Automatically applied on lightweight models (haiku, flash, mini).

Install with Tessl CLI

npx tessl i github:supercent-io/skills-template --skill prompt-repetition

What are skills?

1.56x

Quality

47%

Does it follow best practices?

Impact

97%

1.56x

Average score across 3 eval scenarios

Optimize this skill with Tessl

npx tessl skill review --optimize ./.agent-skills/prompt-repetition/SKILL.md

SKILL.md

Review

Evals

Evaluation results

100%

58%

Prompt Preprocessing Middleware

Model-aware prompt transformer

Criteria

Without context

With context

Target model list

30%

100%

Reasoning model exclusion

100%

CoT pattern detection

100%

CoT skips repetition

100%

Default 2x repetition

100%

Applied marker added

100%

Marker prevents re-application

100%

Newline separator

100%

80% context ratio

100%

Context overflow reduction

100%

Token estimation method

100%

No padding substitution

100%

Without context: $0.3955 · 1m 38s · 21 turns · 28 in / 6,281 out tokens

With context: $0.5186 · 1m 28s · 21 turns · 24 in / 5,180 out tokens

96%

16%

Accuracy Boost for an Inventory Lookup and Quiz System

Position/index repetition count logic

Criteria

Without context

With context

Position pattern triggers 3x

100%

Position keyword set

80%

100%

MCQ uses 2x

100%

Applied marker present

37%

100%

Newline separator used

100%

Target model only

50%

100%

CoT not repeated

100%

No padding used

100%

Duplicate prevention check

100%

Demo output readable

100%

50%

Without context: $0.4444 · 2m 3s · 22 turns · 28 in / 7,426 out tokens

With context: $0.5834 · 1m 36s · 23 turns · 4,167 in / 5,836 out tokens

97%

33%

Multi-Agent Analytics Pipeline with Prompt Preprocessing

Multi-agent duplicate prevention

Criteria

Without context

With context

Marker-based skip

58%

100%

Marker added on transform

60%

100%

Second agent skips re-application

100%

wrap_llm_call pattern

100%

Lightweight model auto-applied

100%

Non-lightweight model skipped

75%

100%

x-prompt-repetition-applied header/metadata

30%

100%

Pipeline simulation output

100%

CoT agent skipped

62%

No duplicate repetition in output

100%

Without context: $0.5051 · 2m 32s · 25 turns · 80 in / 8,378 out tokens

With context: $0.7524 · 2m 38s · 28 turns · 285 in / 8,834 out tokens

Evaluated: 3 days ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

Prompt Preprocessing Middleware Accuracy Boost for an Inventory Lookup and Quiz System Multi-Agent Analytics Pipeline with Prompt Preprocessing

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.