Talk

Set up real-time voice conversations. Phone calls, voice agents, live speech.

This skill is documentation and setup guidance only — it doesn't install code. Before using it: (1) Be prepared to supply provider API keys (Twilio/Telnyx/ElevenLabs/OpenAI) — treat them as secrets and use least-privilege keys where possible. (2) Public webhook tunnels (ngrok, tailscale, cloudflared) expose a local endpoint — prefer short-lived tunnels or allowlists for incoming calls. (3) The docs mention 'memory injection' (loading MEMORY.md into calls) and cost tracking — avoid loading any sensitive personal data into memories that will be spoken on calls. (4) Confirm billing/cost implications for per-minute voice usage and set inbound allowlists to avoid abuse. (5) Verify provider accounts and webhook URLs before enabling phone call integrations. If you want higher assurance, request a version of the skill from a known source or one that includes code you can review; otherwise the content is coherent but sensitive by nature.

Type: OpenClaw Skill Name: talk Version: 1.0.0 The skill bundle is designed to help users configure real-time voice conversations. It consists primarily of markdown documentation (`SKILL.md`, `config.md`, `providers.md`) providing setup instructions and configuration examples for various voice providers. While these files contain shell commands (e.g., `clawhub install`, `ngrok http`, `tailscale funnel`, `cloudflared tunnel`), these commands are transparently presented within code blocks and are directly related to the legitimate operational requirements of setting up voice services (e.g., installing plugins, creating public webhooks). There is no evidence of data exfiltration, obfuscation, malicious execution, persistence mechanisms, or prompt injection attempts designed to subvert the agent for harmful purposes. The content is aligned with the stated purpose and lacks indicators of intentional malicious behavior.