optimized-translating

name: optimized-translating description: Use when translating game content that requires cost-optimized model routing, linear workflow efficiency, and immediate formatting to avoid multi-loop review cycles context: fork agent: Plan model: sonnet

Optimized Translating

Overview

Optimized translating IS systematic linear chain workflow that eliminates review inefficiencies.

Replace multi-loop review cycles with cost-optimized model routing: sonnet for translation/review, haiku for formatting. Immediate formatting prevents downstream processing problems.

Core principle: Separate concerns with appropriate model assignment. Translation ≠ formatting ≠ review.

Violating the letter of the rules is violating the spirit of the rules.

Routing

Pattern: Chain Handoff: user-confirmation Next: none

Task Initialization (MANDATORY)

Before ANY action, create task list using TaskCreate:

TaskCreate for EACH task below:
- Subject: "[optimized-translating] Task N: <action>"
- ActiveForm: "<doing action>"

Tasks:

1. Validate prerequisites
2. Cleanup source content
3. Extract terminology
4. Fork: Translate (sonnet)
5. Fork: Content review (sonnet)
6. Fork: Format review (haiku)
7. Merge and track progress

Announce: "Created 7 tasks. Starting execution..."

Execution rules:

1. TaskUpdate status="in_progress" BEFORE starting each task
2. TaskUpdate status="completed" ONLY after verification passes
3. If task fails → stay in_progress, diagnose, retry
4. NEVER skip to next task until current is completed
5. At end, TaskList to confirm all completed

Task 1: Validate Prerequisites

Goal: Ensure project setup supports optimized translation.

Verification checklist:

• glossary.json exists and is readable
• style-decisions.json exists
• Target content identified and accessible
• Previous translation progress documented

Verification: All prerequisites checked and documented.

Task 2: Cleanup Source Content

Goal: Prepare source material for efficient processing.

Actions:

• Remove markup artifacts and formatting noise
• Standardize heading structures
• Identify and mark special elements (tables, code blocks)
• Extract metadata for context

Verification: Source content is clean and structured.

Task 3: Extract Terminology

Goal: Validate terminology consistency before translation begins.

Process:

1. Scan content for terminology requiring decisions
2. Cross-reference against existing glossary.json
3. Flag unknown terms for user review
4. Update glossary if new terms approved

Verification: All terminology validated or flagged for review.

Task 4: Fork: Translate (sonnet)

Goal: High-quality translation using cost-appropriate model.

CRITICAL: Use Agent tool with sonnet model for translation work.

Instructions for translation agent:

• Focus purely on semantic accuracy
• Apply glossary terms consistently
• Maintain source structure
• Traditional Chinese (Taiwan usage)
• Do NOT format or review—translation only

Verification: Translation completed, semantically accurate, terminology consistent.

Task 5: Fork: Content Review (sonnet)

Goal: Content quality validation using cost-appropriate model.

CRITICAL: Use Agent tool with sonnet model for content review.

Instructions for review agent:

• Review translation for accuracy and readability
• Verify terminology consistency
• Check cultural adaptation appropriateness
• Flag any semantic issues for correction
• Do NOT format—content review only

Verification: Content reviewed, issues flagged or corrected.

Task 6: Fork: Format Review (haiku)

Goal: Structure and formatting using cost-optimized model.

CRITICAL: Use Agent tool with haiku model for formatting work.

Instructions for format agent:

• Apply Starlight markdown formatting
• Convert page references to internal links
• Ensure heading hierarchy compliance
• Apply Traditional Chinese punctuation rules
• Validate frontmatter structure

Verification: Formatting complete, structure validated.

Task 7: Merge and Track Progress

Goal: Integrate results and update progress tracking.

Actions:

• Merge translation, content fixes, and formatting
• Update progress tracker with completion status
• Create summary of work completed
• Prepare handoff documentation

Verification: Results merged, progress updated, ready for user review.

Red Flags - STOP

These thoughts mean you're rationalizing. STOP and reconsider:

• "Skip the fork steps, do it all at once"
• "Sonnet for formatting is fine, costs don't matter"
• "Multiple review rounds will improve quality"
• "Combine translation and review into one step"
• "Skip terminology extraction, it's obvious"
• "Format later, focus on translation first"

All of these mean: You're about to recreate the inefficiencies this skill prevents.

Common Rationalizations

Workflow Pattern

digraph optimized_translation {
    rankdir=TB;

    start [label="Translation\nrequest", shape=doublecircle];
    validate [label="Task 1:\nValidate prereqs", shape=box];
    cleanup [label="Task 2:\nCleanup content", shape=box];
    terms [label="Task 3:\nExtract terminology", shape=box];
    translate [label="Task 4:\nFork: Translate\n(sonnet)", shape=box, style=filled, fillcolor="#e6ffe6"];
    review [label="Task 5:\nFork: Content review\n(sonnet)", shape=box, style=filled, fillcolor="#e6ffe6"];
    format [label="Task 6:\nFork: Format review\n(haiku)", shape=box, style=filled, fillcolor="#ffe6e6"];
    merge [label="Task 7:\nMerge & track", shape=box];
    done [label="Translation\ncomplete", shape=doublecircle];

    start -> validate;
    validate -> cleanup;
    cleanup -> terms;
    terms -> translate;
    translate -> review;
    review -> format;
    format -> merge;
    merge -> done;
}

References

• references/workflow-pattern.md - Detailed translation flow
• templates/progress-tracker.json - Progress state template