Advanced AI voice assistant for phone calls. Capable of persuasion, sales, restaurant bookings, reminders, and notifications.
Security Analysis
medium confidenceThe skill's requested binaries, environment variables, and runtime behavior match its stated purpose (making autonomous phone calls via Vapi), but it requires exposing a local webhook to the internet and writes local logs—so treat it as an internet-facing service and run it in isolation.
The skill is an AI telephony assistant and requests a Vapi API key, assistant ID, phone-number ID, and a public webhook URL; these are appropriate and expected for making/receiving telephony callbacks. Required binary (python3) and requests dependency are proportional. Minor metadata inconsistency: skill name in the registry input ('zzz') doesn't match SKILL.md/package (vapi-calls), and package.json lists envs but registry primary credential is 'none' despite VAPI_API_KEY being required.
SKILL.md and the included script explicitly instruct the agent to start an HTTP server and require the host machine be reachable from the internet (Cloudflare Tunnel/ngrok). The runtime code only reads the declared environment variables (plus optional LLM provider/model), calls the documented API endpoint (https://api.vapi.ai/call), handles webhooks for the expected call ID, and writes per-call JSON logs under ~/.openclaw/workspace/logs. No instructions to read other system files or unrelated credentials were found, but the requirement to expose a local port is a meaningful operational risk that users must evaluate.
This is instruction-only with a small Python script bundled. No external download URLs or archive extraction are used; package.json runs a chmod on the script during postinstall. No high-risk install mechanism detected.
Required env vars (VAPI_API_KEY, VAPI_ASSISTANT_ID, VAPI_PHONE_NUMBER_ID, WEBHOOK_BASE_URL) are relevant to telephony and appear justified. Optional VAPI_LLM_PROVIDER and VAPI_LLM_MODEL are reasonable. One minor inconsistency: the registry metadata lists no primary credential while the skill clearly depends on a sensitive API key (VAPI_API_KEY).
The skill does not request always:true and does not modify other skills. It spawns an HTTP server bound to 0.0.0.0 on a user-specified port and writes call logs under the user's home directory; this is expected for a webhook-based telephony skill but is a persistence/attack surface consideration (open port, reachable webhook).
Guidance
This skill appears to do what it claims (make automated AI phone calls) and requires the Vapi API key, assistant and phone-number IDs, plus a public webhook URL. Before installing: - Treat the host running this skill as an internet-facing service: prefer an isolated VM/container, not your personal laptop or a machine with sensitive data. - Use a dedicated Vapi API key with the minimum required permissions and be prepared to rotate it if needed. - Limit the exposure of the webhook (use Cloudflare Tunnel or a reverse proxy with access controls) and avoid long-lived public ngrok URLs tied to your primary account. - Confirm the package source/repository (SKILL.md references a GitHub repo) and verify the publisher; the registry metadata has minor inconsistencies (skill name/owner vs. files). - Review logs written to ~/.openclaw/workspace/logs and ensure they don't contain PII you want to avoid storing locally; consider redirecting logs to a controlled location. - Ensure you comply with laws and platform policies for automated calls and consent in your jurisdiction. If you need higher assurance, request the publisher to declare the primary credential and confirm the canonical repository or run the code in a sandboxed environment and audit network traffic.
Latest Release
v1.0.2
Major update: The skill has been reworked and rebranded for advanced AI phone call capabilities. - Migrated from legacy PollyReach scripts to a new system focused on triggering autonomous AI phone calls. - Old message scripts (send/receive/NewChatID) and their flows were removed. - Introduced a new tool (make_vapi_call) allowing API-triggered phone calls with customizable instructions and agent behavior. - Added new configuration requirements for secure webhook communication and environment variables. - Updated documentation to reflect new features, usage, and troubleshooting steps.
More by @MrsZ
Published by @MrsZ on ClawHub
