telnyx-agentic
Installation
SKILL.md
Telnyx Agentic
Use this skill for Telnyx-powered outbound voice work where the agent must actually place calls, navigate phone menus, run multi-lead campaigns, recover from Telnyx/API errors, poll transcripts, and prove what happened.
Core Rules
- Do not place an outbound call, send SMS, submit a form, or contact a third party unless the current user explicitly asked to send, fire, call, or otherwise execute that outbound action.
- Never write API keys, tokens, cookies, SIP passwords, or full secrets into repos, skill files, logs, screenshots, or final responses.
- Use
TELNYX_API_KEYfrom the user's existing global environment first. If absent, check common local env loaders such as~/.envwithout printing secret values. - If a call can spend money, verify account balance and outbound destination restrictions before retrying.
- Treat Telnyx conversation transcripts, call states, recordings, and saved JSON logs as proof. Do not claim a call reached a person, left voicemail, or completed from the initial queued response alone.
- Prefer small direct shell/Python/curl commands for one-off operations. Create repo-local reusable wrappers only when the user wants repeatable agentic calling.
- For 2+ leads, use campaign discipline: a lead list, a call plan, per-lead status, retry windows, transcript proof, and a final outcome table. For 10+ leads, use Telnyx Missions or an equivalent state ledger.
- Create or update only campaign-specific assistants. Do not mutate unrelated shared assistants, production routing, TeXML apps, SIP connections, or phone-number assignments unless the user explicitly asked for that infrastructure change.
Choose The Right Telnyx Tool
Do not blindly use every Telnyx feature. Use the smallest set that fits the job.