Skills pine-voice

Give your agent a real phone. It dials, waits on hold, negotiates your bills, and returns a full transcript.

install
source · Clone the upstream repo
git clone https://github.com/openclaw/skills
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/bojieli/pine-voice" ~/.claude/skills/openclaw-skills-pine-voice && rm -rf "$T"
OpenClaw · Install into ~/.openclaw/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.openclaw/skills && cp -r "$T/skills/bojieli/pine-voice" ~/.openclaw/skills/openclaw-skills-pine-voice && rm -rf "$T"
manifest: skills/bojieli/pine-voice/SKILL.md
source content

Pine Voice

Make real phone calls via Pine AI's voice agent. The agent calls the specified number, navigates IVR systems, handles verification, conducts negotiations, and returns a full transcript.

Authentication

Credentials persist in

~/.pine-voice/credentials.json
— users only need to authenticate once.

Before making calls, check if already authenticated:

node {baseDir}/scripts/auth-check.mjs

If

authenticated
is
true
, skip straight to How to make a call. If
false
, run the auth flow below. Ask the user for their Pine AI account email (sign up at https://19pine.ai).

Step 1: Request verification code

node {baseDir}/scripts/auth-request.mjs "user@example.com"

Returns

{"request_token": "...", "email": "..."}
. Save the
request_token
.

Tell the user: "A verification code has been sent to your email. Check your inbox (and spam) and give me the code."

Step 2: Verify and save credentials

node {baseDir}/scripts/auth-verify.mjs "user@example.com" "REQUEST_TOKEN" "CODE"

Returns

{"status": "authenticated", "credentials_path": "..."}
. Credentials are saved automatically.

When to use

Use this skill when the user wants you to make a phone call on their behalf.

Important: The voice agent can only speak English. Supported countries: US/CA/PR (+1), UK (+44), AU (+61), NZ (+64), SG (+65), IE (+353), HK (+852).

Best for

  • Calling customer service to negotiate bills, request credits, or resolve issues
  • Scheduling meetings or appointments by phone
  • Making restaurant reservations
  • Calling businesses to inquire about services or availability
  • Following up with contacts on behalf of the user

How to make a call

Step 1: Gather all required information

Before calling, you must collect every piece of information the callee might need. The voice agent cannot ask a human for missing information during the call. Anticipate what will be required: authentication details, payment info, negotiation targets, relevant context.

Step 2: Initiate the call

Pass call parameters as JSON via stdin:

node {baseDir}/scripts/call.mjs <<'EOF'
{
  "dialed_number": "+14155551234",
  "callee_name": "Comcast Customer Service",
  "callee_context": "Cable and internet provider. Account holder: Jane Doe, account #12345.",
  "call_objective": "Negotiate monthly bill down to $50/mo. Do not accept above $65/mo.",
  "detailed_instructions": "Mention 10-year customer loyalty. If no reduction, ask for retention department.",
  "caller": "negotiator",
  "voice": "female",
  "max_duration_minutes": 60,
  "enable_summary": false
}
EOF

Returns

{"call_id": "..."}
. The call is now active.

Step 3: Poll for results

Poll every 30 seconds until

is_terminal
is
true
:

node {baseDir}/scripts/call-status.mjs "CALL_ID"

When complete, the response includes

transcript
,
duration_seconds
, and
credits_charged
. The
is_terminal
field tells you when the call is done.

IMPORTANT: Use

sessions_spawn
to run this in a background sub-agent so you remain available to the user during the call (which can take 5-60+ minutes).

Example task for sessions_spawn:

Make a phone call using the Pine Voice scripts. Run: node {baseDir}/scripts/call.mjs with stdin JSON: {"dialed_number": "+14155551234", "callee_name": "The Restaurant", "callee_context": "Italian restaurant, making a dinner reservation", "call_objective": "Reserve a table for 4 at 7pm tonight", "caller": "communicator"}. Then poll with: node {baseDir}/scripts/call-status.mjs "CALL_ID" every 30 seconds until is_terminal is true. Report the full transcript and outcome.

Step 4: Evaluate the transcript

Do NOT rely on the

status
field to judge success. Read what the OTHER party actually said.

Treat the call as a FAILURE if:

  • Only Pine's agent speaks and the other side is silent
  • The other party's responses are automated/recorded (voicemail, IVR-only)
  • Extended silence from both sides
  • The callee hung up before the objective was discussed

Call parameters

ParameterRequiredDescription
dialed_number
YesPhone number in E.164 format (e.g.
+14155551234
)
callee_name
YesName of the person or business
callee_context
YesAll context the agent needs: who they are, auth details, verification info
call_objective
YesSpecific goal with targets and constraints
detailed_instructions
NoStrategy, approach, behavioral instructions
caller
No
"negotiator"
(default) or
"communicator"
voice
No
"male"
or
"female"
(default:
"female"
)
max_duration_minutes
No1-120 (default: 120)
enable_summary
No
true
/
false
(default:
false
)

Negotiation calls

For negotiations, set

caller
to
"negotiator"
and provide a thorough strategy:

  • Target outcome: "Reduce monthly bill to $50/mo"
  • Acceptable range: "Will accept up to $65/mo"
  • Hard constraints: "Do not change plan tier"
  • Leverage points: "10-year customer, competitor offers $45/mo"
  • Fallback: "Request one-time credit of $100"
  • Walk-away: "Ask for retention department"

Examples

Test call: "Call my phone at +1XXXXXXXXXX. Tell me that Pine Voice is set up and working."

Restaurant reservation: "Call +14155559876 and make a reservation for 4 tonight at 7pm. If unavailable, try 7:30 or 8pm. Name: Jane Doe."

Model requirements

Pine Voice works best with models that have thinking/reasoning capabilities.

  • Recommended: Claude Sonnet/Opus 4.5+, GPT-5.2+, Gemini 3 Pro
  • Not recommended: Gemini 3 Flash, or models without thinking capabilities

Privacy

Pine Voice processes call data on Pine AI's servers. Credentials are stored locally in

~/.pine-voice/credentials.json
with restricted permissions (600). Call transcripts are returned in the API response and are not stored locally. See https://www.19pine.ai/page/privacy-policy for Pine AI's data handling policies.