| name | whatsapp-personal-channel |
| description | Operate the local unofficial WhatsApp personal channel bridge for a personal account, limited to the allowlisted phone number or numbers in the bridge config. |
| metadata | {"author":"oneshotprompt","version":"1.0"} |
Use this skill when the task is specifically about reading WhatsApp messages, replying to them, or carrying out a clear local machine action requested through the local bridge in tools/whatsapp-personal-channel.
This skill is for a single reply pass after a trigger has already fired. If the operator wants continuous waiting for new messages, the runtime should use the external listen command instead of implementing a loop inside the agent.
Workflow:
- Verify the bridge is running with
RunCommand using fileName=node and arguments=tools/whatsapp-personal-channel/channel.mjs health.
- Use the current trigger details from the task prompt when present, then read unread inbound messages with
RunCommand using fileName=node and arguments=tools/whatsapp-personal-channel/channel.mjs list-unread --limit 10 to confirm the sender and message body.
- If more context is needed, read recent history for that sender with
RunCommand using fileName=node and arguments=tools/whatsapp-personal-channel/channel.mjs list-recent --phone <digits> --limit 10.
- If the inbound message requests a clear and safe local machine action, load the
command-execution skill and use the narrowest direct command needed to perform it.
- After the action succeeds, send exactly one concise confirmation to that same sender with
RunCommand using fileName=node and arguments=tools/whatsapp-personal-channel/channel.mjs send --phone <digits> --text "...".
- If the request is ambiguous, risky, or the local command fails, send exactly one concise clarification or failure message to that same sender with
--phone <digits>.
- Summarize what actually happened after the command sequence: no unread messages, waiting for bridge readiness, a local action that was completed, or the reply that was sent.
Guardrails:
- Do not attempt to message any number outside the bridge allowlist.
- The bridge already enforces
allowedPhoneNumbers; treat any allowlist error as a hard stop.
- If the bridge is not ready, report that status instead of guessing.
- Always pass
--phone <digits> when using send or list-recent so the reply is pinned to the correct sender.
- Do not send more than one reply in a single run unless the task explicitly says to do so.
- Prefer short, direct replies because this runs against a personal WhatsApp account.
- Only execute local actions when the request is explicit and the command is safe to run with the tools already selected for the job.
- For destructive, risky, credentialed, or unclear requests, ask for confirmation instead of acting.
- Never claim a local action succeeded unless the exit code and command output confirm it.
- If
list-unread returns no messages but the current trigger clearly identifies the sender and request, you may still use that trigger context for this run.
- Do not build your own polling or waiting loop inside the agent; let the CLI
listen mode handle continuous waiting.
Command behavior:
health returns JSON describing readiness and the current linked account name.list-unread returns JSON with unread inbound messages only from allowlisted numbers.list-recent returns recent messages for the allowlisted number. Pass --phone <digits> to target the intended sender.send sends a message only to an allowlisted number. Pass --phone <digits> so the reply always goes back to the intended sender.
