bilingual-content-sync-agent 0.4.0

Solo Founder OS agent #4 — keep EN/ZH locale JSON in sync via Claude + HITL markdown review.

bilingual-content-sync-agent

English | 中文

Solo Founder OS agent #4 — keep en.json and zh.json (or any locale pair) in sync. Diffs, translates missing keys via Claude, queues them in a markdown HITL review file, applies approved edits back to your locale JSON. Never auto-writes without human approval.

Built by Alex Ji for VibeXForge's 925 EN/ZH i18n strings. Born from this thought:

Every feature ships English first. ZH catches up days or weeks later. Tracking which keys are stale is manual labor I do at 1 AM. Same for the next 11 makers I know.

What it does

en.json + zh.json
        ↓
   diff (find keys present in EN, missing/empty/echo'd in ZH)
        ↓
   translate via Claude Sonnet (vibex-tone glossary baked in)
        ↓
   queue/pending/<timestamp>-review.md
        ↓
   founder edits in Obsidian, moves to queue/approved/
        ↓
   apply → writes back to zh.json (preserves nesting + key order)

CLI

# Coverage report
bilingual-sync stats --en locales/en.json --zh locales/zh.json

# Show what's missing (no LLM call)
bilingual-sync diff --en locales/en.json --zh locales/zh.json

# Translate + write review file (LLM call)
bilingual-sync draft \
  --en locales/en.json \
  --zh locales/zh.json \
  --glossary glossary.json \
  --tone "indie SaaS, gamified, neon-pixel, peer-to-peer"

# After human review → move file to queue/approved/, then:
bilingual-sync apply --en locales/en.json --zh locales/zh.json

Glossary file

Glossary keys force consistent translation of brand terms. Without it, Claude picks reasonable defaults but those drift across runs.

{
  "Forge": "锻造",
  "Dojo": "道场",
  "Vibe": "氛围",
  "Login": "登录",
  "Sign up": "注册"
}

What gets flagged as "missing"

A key is missing in ZH when ANY of:

• key absent from zh.json
• ZH value is empty string
• ZH value byte-equals the EN value (a common copy-paste mistake — easy to miss without tooling)

Design choices

• HITL is non-negotiable. UI strings are seen by every user every session. One bad translation that ships globally is worse than five days of catch-up. The agent never writes to zh.json without an explicit approve-and-apply step.
• Sonnet by default, not Haiku. Cold UI translations are exactly the task where naïve Haiku output gets too literal ("Login" → "登入" instead of "登录"). Sonnet's nuance is worth the 3× cost for 925 strings (~$0.50 total).
• Batched calls. 50 keys per call is the token-budget sweet spot. Full vibex catalog = ~20 calls = ~$0.50.
• Preserves placeholders + HTML. The system prompt explicitly forbids translating {name}, <span>, ICU patterns. If Claude breaks this it gets flagged in the review file.
• Built on solo-founder-os. AnthropicClient handles graceful degrade + auto-logs token usage to ~/.bilingual-content-sync-agent/usage.jsonl (cost-audit-agent picks it up monthly).

Roadmap

• v0.1 — diff/draft/apply/stats CLI · HITL queue · Sonnet drafting · 43 tests · graceful degrade
• v0.2 — Multi-language support (EN→ES, EN→JA, etc.) — same code path, different glossaries
• v0.3 — Watch mode: re-run on commit, post-flagging stale keys to GitHub PR
• v0.4 — In-context preview: show the React component a key appears in
• v0.5 — Reverse sync: detect ZH-only orphan keys (features that got removed but ZH was forgotten)

MCP server (Claude Desktop / Cursor / Zed)

Translate missing i18n keys inline from your AI assistant.

pip install 'bilingual-content-sync-agent[mcp]'

{
  "mcpServers": {
    "bilingual-sync": {
      "command": "bilingual-sync-mcp",
      "env": { "ANTHROPIC_API_KEY": "..." }
    }
  }
}

Tools: diff_locales(en, zh) · translate_missing(en, zh, ...) · apply_approved(en, zh)

License

MIT.

🧩 Part of the Solo Founder OS stack

A growing collection of MIT-licensed agents that share solo-founder-os as their base — cron, eval, reflexion, AnthropicClient, HITL queue, notifiers. Each agent is independently useful; together they cover the full one-person company workflow.

🌐 The whole stack is live in production at vibexforge.com.

Agent	What it does
solo-founder-os	The shared base lib (cron · eval · reflexion · skill library · DGM-lite). Every other agent depends on it.
build-quality-agent	Pre-push Claude diff reviewer + local build runner — catches CI-killing changes before they ship.
customer-discovery-agent	Reddit pain-point scraper + Claude clustering for product validation.
funnel-analytics-agent	Daily founder brief + real-time PH-day alerts across 9 sources.
orallexa-marketing-agent	AI marketing agent — submit project once, get platform-native posts for X / Reddit / HN / Dev.to / 小红书 + 7 more. Powers vibexforge.com.
vc-outreach-agent	Cold email drafter — investors (vc mode) or paying customers (customer mode, merged from customer-outreach in v0.9.0). HITL queue + SMTP sender.
cost-audit-agent	Monthly bill audit across 6 providers (Vercel / Anthropic / OpenPanel / HyperDX / Supabase / GitHub Actions) with dollar-tagged waste findings.
customer-support-agent	Triage user messages → auto-draft replies → HITL queue. Closes the L5 customer-support layer.
payments-agent	Overdue-invoice reminder drafter. Stripe-shaped types + MockProvider fallback + hard money-safety prompt rules.

Each agent's own row is omitted from its README. Install whichever solve real problems for you — pip install <agent-name>.