Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add skills-il/developer-tools --skill hebrew-voice-bot-buildergit clone --depth 1 https://github.com/skills-il/developer-toolsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/skills-il/developer-tools/hebrew-voice-bot-builder)<a href="https://agentmods.dev/skills/skills-il/developer-tools/hebrew-voice-bot-builder"><img src="https://agentmods.dev/badge/skills/skills-il/developer-tools/hebrew-voice-bot-builder/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/skills-il/developer-tools/hebrew-voice-bot-builder"><img src="https://agentmods.dev/badge/skills/skills-il/developer-tools/hebrew-voice-bot-builder.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Memory Poisoning · line 30 Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.Fix: Implement context-window management that detects and rejects padding or stuffing attempts. Prioritize system instructions over user-injected content.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00192 | $0.09866 |
| Opus 5.5 | $0.00077 | $0.03946 |
| Sonnet 5.5 | $0.00038 | $0.01973 |
| Haiku 4.5 | $0.00019 | $0.00987 |
Grade A, and why
hebrew-voice-bot-builder scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 16d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 705 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hebrew Voice Bot Builder
Build production-ready Hebrew voice bots and IVR systems for Israeli businesses. This skill covers the full voice pipeline: speech-to-text (STT), text-to-speech (TTS), IVR flow design, telephony integration, and Hebrew-specific challenges like accent handling and mixed Hebrew-English speech.
Instructions
Language code: Google STT uses
iw-IL, Google TTS useshe-IL. Google's Speech-to-Text supported-languages table lists Hebrew asiw-IL(the legacy ISO code for Hebrew), while the Text-to-Speech voice list useshe-IL(he-IL-Wavenet-A,he-IL-Chirp3-HD-*). Pass the documented code for each side rather than assuming one code works for both.
Step 1: Choose Your Architecture
Before building, decide on the voice bot architecture based on the use case:
| Architecture | Best For | Components |
|---|---|---|
| IVR (keypad) | Simple menu navigation, payment lines, appointment scheduling | TTS + DTMF + telephony |
| Voice bot (conversational) | Customer service, order status, FAQ handling | STT + LLM + TTS + telephony |
| Voicemail transcription | Missed call handling, message routing | STT + notification pipeline |
| Hybrid | Complex flows with both speech and keypad input | STT + TTS + DTMF + telephony |
Key decisions:
- STT provider: OpenAI
gpt-transcribefor recorded audio andgpt-live-transcribefor streaming (Realtime transcription sessions only, not the file endpoint). On 2026-08-26 OpenAI deprecatedwhisper-1,gpt-4o-transcribe,gpt-4o-mini-transcribeandgpt-4o-transcribe-diarize, all shutting down on 2027-02-26, so do not start new work on them.whisper-large-v3-turbofor self-host, ivrit-ai's Hebrew-tuned variants (ivrit-ai/whisper-large-v3-turbo-ct2) for an open model trained specifically on Hebrew, Google Cloud STT (low latency; Hebrew runs on Chirp,iw-IL), Azure Speech (enterprise features), and ElevenLabs Scribe v2, which lists Hebrew (heb) in the "Good (>10% to <=20% WER)" band with a realtime variant at roughly 150ms. Note the WER band honestly: Hebrew is two tiers below English there, so benchmark on your own call audio rather than picking on the marketing claim. - TTS provider, split by use case:
- Real-time / streaming (voice agents, IVR, live conversation): OpenAI Realtime API, native multilingual speech-to-speech including Hebrew over WebRTC/WebSocket/SIP, the 2026 default for sub-500ms turn-taking. The current GA model is
gpt-realtime-2.1(gpt-realtime-2.1-minifor the smaller tier). Two model names to avoid:gpt-realtimewas deprecated on 2026-07-20 with a hard shutdown on 2027-01-20 (replacementgpt-realtime-2.1), andgpt-4o-realtime-previewwas removed from the API on 2026-05-07.gpt-realtime-1.5is still live. Other realtime options: ElevenLabseleven_v4_turbo(~100ms, Hebrew is in the v4 language list; ElevenLabs now labelseleven_v3_conversationalprevious generation), Inworldinworld-tts-2andinworld-tts-2-flash(Hebrewheis listed as a Tier 1 language; the 1.5 models are deprecated), and the Israeli vendor Deepdub (API model iddd-etts-3.0, Hebrewhe-ILin its API language table). Do NOT reach for ElevenLabseleven_flash_v2_5for Hebrew: its language list is Multilingual v2's 29 languages plus Hungarian, Norwegian and Vietnamese, so Hebrew is absent entirely rather than merely weak. - Offline / max quality (audiobooks, voicemail playback, batch generation): ElevenLabs
eleven_v4(Hebrew is in its 90+ language list;eleven_v3is now previous generation). v3 and v4 are NOT on the Text to Speechstream-inputWebSocket; v4 is served through the Text to Dialogue API, and their WebSocket route is the Text to Dialogue WebSocket (wss://api.elevenlabs.io/v1/text-to-dialogue/stream-input). Deepdub also serves this track with emotional control. - Fallbacks: Azure Neural TTS (
he-IL-HilaNeural,he-IL-AvriNeural), Google Cloud TTS Wavenet (he-IL-Wavenet-A/B). Amazon Polly does NOT support Hebrew (no he-IL locale, no Hebrew voice of any engine, the "Avri" voice belongs to Azure, not Polly), so do not route Hebrew through Polly. ElevenLabs Multilingual v2 does NOT list Hebrew: its documented 29 languages are en, ja, zh, de, hi, fr, ko, pt, it, es, id, nl, tr, fil, pl, sv, bg, ro, ar, cs, el, fi, hr, ms, sk, da, ta, uk and ru. Hebrew appears in the Eleven v3 and v4 language lists, so route Hebrew to v4 or v3 (or to Azure/Google above) and not to Multilingual v2.
- Real-time / streaming (voice agents, IVR, live conversation): OpenAI Realtime API, native multilingual speech-to-speech including Hebrew over WebRTC/WebSocket/SIP, the 2026 default for sub-500ms turn-taking. The current GA model is
- Telephony: Twilio and Vonage both sell Israeli numbers. Twilio publishes Israeli voice pricing (local, mobile and toll-free tiers); Vonage does not publish comparable Israeli number documentation, so compare quotes yourself rather than trusting a ranking. Number portability exists in the Israeli market, but confirm with the carrier that your specific number can be ported to your chosen provider before committing.
- Hosting: Cloud functions for low-volume, dedicated servers for high-volume.
- Call recording and voiceprints: play a recording announcement (
השיחה מוקלטת) at the start of the call. It is standard practice, not a cited statutory duty. The Wiretap Law (חוק האזנת סתר) defines "האזנת סתר" as "האזנה ללא הסכמה של אף אחד מבעלי השיחה". Treat a voiceprint as a different artifact from the recording: store it in its own table, keyed for per-caller deletion, so it can be erased independently of the audio and of the transcript. Have the operator confirm which obligations apply to their business with a qualified professional before launch.
What ships with it
10 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- evidence.json 45 KB
- metadata.json 1.5 KB
- optimization-log.json 8.1 KB
- references/hebrew-stt-models.md 9.2 KB
- references/hebrew-tts-normalization.md 9.5 KB
- references/ivr-design-patterns.md 11 KB
- references/twilio-call-flow.md 16 KB
- scripts/hebrew-stt-demo.py 14 KB runs code
- scripts/test_hebrew_tts_normalization.py 9.4 KB runs code
- SKILL_HE.md 44 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 16d ago Changed · -106 lines 8f472d478376
- 1mo ago First seen · 811 lines · 192 tokens per session scan A 58af21d4b3b5
hebrew-voice-bot-builder is a skill published in the GitHub repository skills-il/developer-tools (13 stars, last pushed 3d ago), licensed MIT. It adds 192 tokens to every session and 9,866 once invoked, about $0.0008 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
heygen
Prepare guarded HeyGen Direct API requests for avatar video generation, video translation, asset discovery, and status polling.
contentful
You are an expert in Contentful, the API-first content platform for enterprise teams. You help developers integrate Contentful's Content Delivery API (CDN-backed, read), Content Management API (write), and Content Preview API (draft content) into websites and apps — using typed content models, localization, rich text…
webiny-v5-to-v6-migration
Migration patterns for converting v5 Webiny code to v6 architecture. Use this skill when migrating existing v5 plugins to v6 features, converting context plugins to DI services, adapting v5 event subscriptions to v6 EventHandlers, or understanding how v5 patterns translate to v6. Targeted at AI agents performing…
contentful-cms
Contentful content types, GraphQL/REST queries, CLI migrations, assets and locales. Use when changing a Contentful content model, querying Contentful, or migrating its content.
email-template-builder
Construtor de templates de e-mail transacional. Cria sistemas completos de e-mail com React Email, integração com provedores, servidor de preview, suporte a i18n, modo escuro, otimização anti-spam e rastreamento de analytics. Produz código pronto para produção para Resend, Postmark, SendGrid ou AWS SES.
backend-internationalization
Use this skill when implementing multi-language support, translation workflows, or locale-aware formatting. This skill enforces: BCP 47 locale tags, ICU MessageFormat for complex messages, dot-separated namespaced keys, fallback chains, and CI-validated translation files. Applies to any backend stack serving…