feat: add Hindi language support to i18n module

[tejasashinde]

Add Hindi language support to i18n module

This PR adds Hindi localization support via a new hi.json file, extending the system to support Devanagari-based Hindi language output for CLI and UI messages.

FEATURES

1. Core terminology

Hindi translations for main system terms:

• महल, खंड, हॉल, अलमारी, दराज़, खनन, खोज, स्थिति, आरंभ, मरम्मत, स्थानांतरण, इकाई, विषय

2. CLI localization

• Full Hindi translations for all CLI messages
• Proper grammar handling (e.g. अलमारियाँ बनाई गईं, दराज़ें बनाई गईं)

3. AAAK instruction

• Hindi version of compression rules:
  • Index-based formatting
  • - for word separation, | for concept separation
  • Preserve names and numbers exactly

4. Regex (Hindi support)

• Devanagari topic detection: [\\u0900-\\u097F]{2,}
• ASCII fallback for identifiers and CLI tokens
• Hindi stop-word filtering (eg. यह वह ये वे कुछ कई अधिकांश प्रत्येक हर अन्य केवल ऐसा बहुत होगा सकता चाहिए etc.)
• Language-agnostic action pattern kept unchanged

NOTES

• No code changes are required
• Auto-loaded via i18n/*.json glob pattern
• Fully aligned with existing locales (en, es, fr, de, ja, ko, zh-CN, zh-TW).

[igorls]

Thanks @tejasashinde! Schema matches en.json, interpolation variables ({path}, {count}, {closets}, {drawers}, {fixed}, {query}) all correct, topic_pattern properly uses the Devanagari block [\u0900-\u097F]{2,} with an ASCII fallback. Nice.

One blocker before merge: action_pattern is still the English verb list (built|fixed|wrote|added|pushed|measured|tested|reviewed|created|deleted|updated|configured|deployed|migrated). The PR description says this is "kept unchanged" as language-agnostic, but every other locale (es, fr, de, ja, ko, zh-CN, zh-TW, and the recently-landed ru/it/id) localizes it — this pattern is used to extract "what the user did" from prose, so for Hindi text it'll silently match nothing.

Please add Devanagari equivalents — something like:

(?:बनाया|सुधारा|लिखा|जोड़ा|भेजा|मापा|परीक्षण किया|समीक्षा की|सृजित किया|हटाया|अपडेट किया|कॉन्फ़िगर किया|तैनात किया|स्थानांतरित किया)\\s+[\\w\\s]{3,30}

(double-check with your own Hindi ear — those are my best guesses from the English verbs.)

Also, could you trigger CI so we can confirm everything's green? No checks have run on this branch yet.

Optionally — #911 just landed infra for Hindi-aware entity detection. You can add an entity section to hi.json with Hindi person verbs (कहा, पूछा, बोला, etc.), pronouns (वह, उसने, उन्होंने), and a Devanagari candidate_pattern so Hindi names like राज, अनीता get extracted from prose. Totally optional; CLI/AAAK Hindi is a standalone win.

[tejasashinde]

Hi @igorls, thanks for pointing that out. This is now fixed.

I’ve updated hi.json with the correct action_pattern and also added support for the new infra as per #911, including entity, pronoun_patterns, dialogue_patterns, direct_address_pattern, project_verb_patterns, and stopwords.

I noticed CI didn’t trigger on this fork PR, and I didn’t see a CI workflow configured in this repository. I ran the tests locally using pytest as per the contributing guide:

python -m pytest tests/test_i18n.py -v
python -m pytest tests/test_entity_detector.py -v
python -m pytest tests/test_readme_claims.py -v

Please let me know if anything else needs to be done from my side.

[igorls]

@tejasashinde — great news: #932 just landed, which fixes the \b word-boundary issue for Devanagari combining marks (matras). Your entity section is structurally correct but needs one addition to activate it.

Add boundary_chars to the entity section in hi.json:

"entity": {
  "boundary_chars": "\\w\\u0900-\\u097F",
  "candidate_pattern": "[\\u0900-\\u097F]{2,20}",
  ...

That single field tells the i18n loader to replace every \b in Hindi patterns with a script-aware boundary that treats Devanagari vowel signs (ा ी ु ू etc.) as inside-word characters. Without it, Python's \b splits on matras — names like अनीता (Anita) truncate to अनीत, and person-verb patterns like \bराज\s+ने\s+कहा\b never match because the ा at the end of कहा isn't a \w character.

Also, one small fix in regex.action_pattern: replace [\\w\\s] at the end with [\\w\\s\\u0900-\\u097F] so the AAAK extraction captures Devanagari content after action verbs:

"action_pattern": "(?:बनाया|सुधारा|...)\\s+[\\w\\s\\u0900-\\u097F]{3,30}"

Once those two changes land and CI is green, this is ready to merge!

[tejasashinde]

@igorls - Done!

Summary

Fix Hindi regex matching by making boundaries Devanagari-aware.

Changes

• Added boundary_chars to entity in hi.json
• Updated action_pattern to include Devanagari range ([\w\s\u0900-\u097F])

Testing

Re-ran tests locally:

python -m pytest tests/test_i18n.py -v
python -m pytest tests/test_entity_detector.py -v
python -m pytest tests/test_readme_claims.py -v