Turkish Native v0.5.0

Turkish Native v0.5.0

v0.5.0 is the first tagged release after v0.3.0. It includes the production work that landed on main as v0.4.0 plus a new domain and page integrity layer.

Highlights

• 57 Turkish-specific patterns, up from 48 in v0.3.0.
• 105 regression cases, up from 72.
• Set-level tests for heading monotony and duplicate meaning.
• New page-integrity patterns for cross-sector template residue, domain-native fields and actions, unsupported operational promises, high-stakes claims, provenance/source labels, and process-section theater.
• Positive controls that preserve context-sensitive language when it is actually correct.
• A privacy regression case for unsupported absolute data-handling claims.

Why this release exists

Recent production reviews surfaced several examples that were grammatically fine but contextually wrong: a florist CTA asked users to book an appointment, an auto-service form asked users to choose a vehicle segment, a restaurant used inventory language such as stok, a generated dental demo made unsupported outcome claims, synthetic placeholder reviews were presented as Google reviews, and obvious UI actions were expanded into artificial multi-step processes.

Turkish Native now treats sector, task, provenance, and factual scope as part of writing quality rather than as separate cleanup work.

Installation

npx skills add oguzhankayan/turkish-native --global

The core principle has not changed: work from meaning and context, preserve supported facts, and do not replace one template-shaped failure with another.

Turkish Native v0.3.0 — OSS preview

Turkish Native v0.3.0 — OSS preview

First public-ready preview of Turkish Native, an agent skill for writing Turkish from meaning instead of translated-English sentence structure.

Highlights

• 48 Turkish-specific writing and localization patterns.
• 72 regression cases with full pattern coverage.
• Positive controls to prevent overcorrection of already-natural Turkish.
• English and Turkish documentation.
• Claude plugin packaging and OpenAI-compatible agent metadata.
• Package validation and GitHub Actions CI.
• Contribution guide, security policy, code of conduct, issue templates, and pull request template.

What the skill targets

Turkish Native focuses on Turkish that is grammatically understandable but still sounds translated: imported clause structures, nominalization chains, wrong case or direction choices, unnatural verb–noun pairings, literal SaaS and sales calques, internal product jargon, forced slogans, repetitive model-written copy, and register mismatches.

It is deliberately conservative. Natural Turkish should stay natural, and supported facts, numbers, names, citations, product terminology, code, keys, URLs, and identifiers should not be invented or silently changed.

Status

This is a pre-1.0 release. Pattern names, evaluation schema, packaging, and benchmark methodology may still change as the corpus grows and more native-reader feedback is collected.

Installation

npx skills add oguzhankayan/turkish-native --global

Claude Code users can also install it as a plugin through the repository marketplace manifest.

Feedback

Bug reports, pattern proposals, positive controls, cross-agent compatibility fixes, and benchmark contributions are welcome through GitHub Issues and pull requests.