translator

name	translator
description	Environmental scanning report Korean translation skill. Translates EN reports to KO and validates quality. Use when — (1) KO report is missing after workflow completion, (2) user requests report translation, (3) KO file assurance is needed before dashboard build. Invoked via /translate command.

Korean Translation Skill

Purpose

Translates English (EN) reports from the environmental scanning system into Korean (KO). Translation quality is validated by translation_validator.py, adhering to the translation-terms.yaml terminology map.

---

When to Use

1. KO report missing after workflow completion: When Step 5.2 (dashboard) detects KO file absence
2. Manual user request: /translate command execution
3. M4 Gate remediation: Automatically invoked on CG-002/004 failure

---

Architecture

┌──────────────────────────────────────────────┐
│  Translator Skill                            │
│                                              │
│  Input: EN report (.md)                       │
│  Terms: translation-terms.yaml (200+ terms)   │
│  Agent: @translation-agent (LLM translation)  │
│  Validate: translation_validator.py (11 checks│
│  Output: KO report (*-ko.md)                  │
└──────────────────────────────────────────────┘

---

Execution Modes

Mode 1: Single File Translation

/translate --file <path-to-en-report.md>

1. Read EN source file
2. Determine output path (*-ko.md suffix convention)
3. Read env-scanning/config/translation-terms.yaml
4. Invoke @translation-agent sub-agent with:
   • Source file path
   • Output file path
   • Terminology map
   • Quality threshold: 0.95
5. Run translation_validator.py on output
6. Report result

Mode 2: All Reports for a Date

/translate --date 2026-03-24

1. Read SOT (workflow-registry.yaml) for enabled workflows
2. For each enabled WF, check if KO report exists
3. Translate ALL missing KO reports (parallel where possible)
4. Validate each KO report
5. Summary report

Mode 3: Auto (called from Step 5.2)

When invoked programmatically from the dashboard step:

1. Receive list of missing KO file paths
2. Translate each (EN source path → KO output path)
3. Validate
4. Return success/failure list

---

Translation Protocol

Step 1: Load Configuration

# SOT binding — all paths are read from SOT
TERMS_FILE = "env-scanning/config/translation-terms.yaml"  # from SOT system.bilingual
VALIDATOR = "env-scanning/core/translation_validator.py"    # from SOT system.bilingual

Step 2: Determine File Pairs (SOT-Driven)

EN/KO filenames are read from SOT (do not hardcode):

# Fields to read from workflow-registry.yaml:
workflows.{wf_key}.deliverables.report_en  # EN filename pattern
workflows.{wf_key}.deliverables.report_ko  # KO filename pattern
workflows.{wf_key}.data_root               # Report directory
integration.deliverables.report_en         # Integrated EN
integration.deliverables.report_ko         # Integrated KO
integration.output_root                    # Integrated directory

Current SOT values (for reference only; always read from SOT at runtime):

WF	report_en	report_ko
wf1-general	environmental-scan-{date}.md	environmental-scan-{date}-ko.md
wf2-arxiv	environmental-scan-{date}.md	arxiv-scan-{date}-ko.md
wf3-naver	naver-scan-{date}.md	naver-scan-{date}-ko.md
wf4-multiglobal-news	environmental-scan-{date}.md	environmental-scan-{date}-ko.md
integrated	integrated-scan-{date}.md	integrated-scan-{date}-ko.md

Step 3: Translate via @translation-agent

Invoke @translation-agent sub-agent for each file:

Instructions to pass to @translation-agent:

1. Read the English source file: {EN_PATH}
2. Read the terminology map: env-scanning/config/translation-terms.yaml
3. Translate the ENTIRE document to Korean following these rules:
   - IMMUTABLE: STEEPs terms (Social, Technological, Economic, Environmental, Political, spiritual), signal IDs, URLs, dates, scores, file paths
   - PRESERVE: arXiv, GPT, AI, EU, WHO, UN and other proper nouns
   - USE MAPPINGS: Follow terminology map (e.g., "weak signal" -> "약한 신호", "environmental scanning" -> "환경 스캐닝")
   - REGISTER: Formal Korean (합쇼체)
   - STRUCTURE: 100% preserve original markdown structure (headings, tables, lists, bold, etc.)
4. Write output to: {KO_PATH}
5. Do NOT add, omit, or reinterpret any content. Translation only.

Step 4: Validate

python3 env-scanning/core/translation_validator.py \
  --source {EN_PATH} \
  --target {KO_PATH} \
  --terms env-scanning/config/translation-terms.yaml

Validation checks:

• STRUCT-001~008: Structure preservation (heading count, table count, list item count, horizontal rules, links, code blocks, total line count ratio +/-40%, blank line ratio)
• TERM-001: STEEPs immutable term preservation
• TERM-002: Proper noun over-translation prevention
• TERM-003: Terminology map compliance >= 60%

Step 5: Retry on Failure

On validation failure:

1. Pass failure item feedback to @translation-agent
2. Re-translate (max 2 retries)
3. After 2 failures, request manual review from user

---

Quality Standards

Criterion	Value	Rationale
Structure preservation rate	>= 90%	Maintain original markdown structure
Terminology map compliance	>= 60%	TERM-003 standard
Korean character ratio	>= 30%	M4 CG-007 standard
Immutable term preservation	100%	STEEPs framework
Translation register	Formal Korean	Official report style

---

Integration Points

Dashboard Step 5.2

Before dashboard generation, checks for KO file existence and invokes this skill if missing.

M4 Gate Remediation

This skill is invoked as a remediation path when CG-002 (KO report exists) fails.

Manual Invocation

Users can directly invoke /translate --date YYYY-MM-DD to supplement missing translations.

---

Dependencies

• Worker Agent: .claude/agents/workers/translation-agent.md
• Validator: env-scanning/core/translation_validator.py
• Terms Map: env-scanning/config/translation-terms.yaml
• SOT: env-scanning/config/workflow-registry.yaml (bilingual section)