All authors
Claude Skills by mahimailabs
github.com/mahimailabs10 skills1 installs7 views
- Voice Agent EvalsTest and evaluate a voice agent. Covers what to assert deterministically versus what to judge with an LLM, the eight judge types, scenario files, simulated callers with accents and background noise, audio versus text runs, regression suites in CI, and a pass-rate maturity curve. Use when someone asks how to test my voice agent, before a release, when a prompt change breaks something silently, or when setting an acceptable pass rate.Votes: 0GitHub stars: 2
- Voice Agent ReviewReview a voice agent codebase or design against a 40-item checklist and produce a scored report covering pipeline choice, turn-taking, interruptions, latency, prompting, tools, telephony, and evals. Each failed item names the skill to read next. Use when someone asks to review my voice agent, asks whether a voice agent is production ready, before a launch, or when auditing an inherited voice agent codebase.Votes: 0GitHub stars: 2
- Voice Full DuplexBuild on full-duplex speech models that listen while they speak. Covers what breaks compared with turn-based models, including no turn detector, no forced stop, tools delegated to a backend model, no verbatim script, no text-only mode, append-only context after startup, side channels, and cost accounting. Use when moving to GPT-Live or another model that listens while it speaks, when a full-duplex model will not stop talking and no cancel call works, when tools never fire, or when say() raises.Votes: 0GitHub stars: 2
- Voice Function ToolsDesign function tools for voice agents. Covers tool descriptions written for spoken intent, confirming arguments before side effects, timeouts and filler speech so the line does not go silent, async and background tools, max tool steps, error strings the agent can say aloud, and context size. Use when adding a tool or function call to a voice agent, when tool calls cause dead air, when the agent books or changes the wrong thing, or when a tool call loops.Votes: 0GitHub stars: 2
- Voice InterruptionsHandle callers who speak while the agent is speaking. Covers barge-in, interruption modes, false interruptions from coughs and backchannels, resuming after a false stop, uninterruptible segments such as the read-back gate before any write, and whether filler speech during tool calls stays interruptible. Use when the agent stops for mm-hm and background noise, when it will not stop when told to, when a confirmation gets cut off, or when tuning barge-in sensitivity.Votes: 0GitHub stars: 2
- Voice Latency BudgetBuild, measure, and fix a latency budget for a voice agent. Covers the five timings that matter (transcription delay, end-of-turn delay, LLM time to first token or TTFT, TTS time to first byte or TTFB, end to end), targets per pipeline shape, where the milliseconds hide, and the order to fix them in. Use when the agent feels slow or laggy, when choosing regions or providers, when reading latency metrics, or when someone asks what good latency is.Votes: 0GitHub stars: 2
- Voice Pipeline ChoiceChoose the pipeline shape for a voice agent, whether cascaded STT-LLM-TTS, a speech-to-speech realtime model, a half-cascade (realtime model in text mode plus a separate TTS), or full-duplex. Covers what each shape gives up on exact wording, latency, prosody, and cost per minute. Use when starting a voice agent or choosing its architecture, when someone asks whether to use the realtime API or speech-to-speech, when a speech-to-speech model paraphrases a number, date, or confirmation code that...Votes: 0GitHub stars: 2
- Voice PromptingWrite system prompts for agents that speak instead of type. Covers spoken-output rules, splitting persona from task, writing numbers, dates, and phone numbers so the TTS says them correctly, confirmation phrasing, sentence length for interruptible speech, and what never belongs in a voice prompt. Use when writing or reviewing a voice agent system prompt, when the agent reads markdown or bullet points aloud, when replies run too long or the agent talks too much, or when the TTS mangles or misp...Votes: 0GitHub stars: 2
- Voice TelephonyPut a voice agent on a phone number. Covers SIP trunks and dispatch, inbound versus outbound, answering machine detection and its five outcomes, DTMF keypad input and IVR phone tree navigation, cold versus warm transfer, narrowband audio, recording consent, and outbound calling rules. Use when connecting an agent to PSTN, SIP, Twilio, or Telnyx, when calls hit voicemail, when transfers drop, when callers sound muffled to the STT, or when planning outbound calls.Votes: 0GitHub stars: 2
- Voice Turn TakingDecide when the caller has finished speaking. Covers VAD, endpointing min and max delay, semantic and audio end-of-turn models, transcription-based turn signals, and user turn limits. Use when the agent keeps interrupting callers or talks over them, waits too long before replying, cuts people off mid-sentence, interrupts someone reading a phone number or date of birth digit by digit, or when tuning turn detection, endpointing, VAD thresholds, or silence timeouts.Votes: 0GitHub stars: 2