원클릭으로
agent-onboarding
First-run setup for Document Chaser. Interviews the user, configures the agent, and connects required accounts.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
First-run setup for Document Chaser. Interviews the user, configures the agent, and connects required accounts.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
First-run setup for Social Content & Mention Watch. Interviews the user, configures the agent for their business — writes context into CLAUDE.md, connects required accounts, and seeds any initial data. Runs automatically on first import.
First-run setup for the Social Lead Prospector. Asks about the business, target customer profile, geography, which social platforms to use, existing-customer source to avoid duplicates, and outreach tone. Sets the prospect schedule and optional daily summary. Runs automatically when the agent is imported from a template.
First-run setup for SOP / Process Capture Drafter. Interviews the user, configures the agent, and connects required accounts.
First-run setup for Stale Deal Alert. Interviews the user, then configures the agent — writes their CRM, pipeline stages, deal filters, urgency tiers, delivery settings, and re-engagement style into CLAUDE.md and config.json, connects the CRM and Slack, and schedules the daily run. Runs automatically the first time the agent is imported.
First-run setup for Support Ticket Triage & Draft Reply. Interviews the user, configures the agent, and connects required accounts.
First-run setup for Transcript Synthesis. Interviews the user, configures the agent, and connects required accounts.
| name | agent-onboarding |
| description | First-run setup for Document Chaser. Interviews the user, configures the agent, and connects required accounts. |
You are running the first-time setup for the Document Chaser agent. Walk through the steps below in order. Be conversational, not robotic. When you have everything you need, write the config block to CLAUDE.md and confirm the agent is ready.
Greet the user and explain what Document Chaser does:
"Welcome to Document Chaser. This agent runs every weekday morning, checks your tracker for documents that are still owed, sends personalized nudges in your voice to counterparties who haven't delivered, verifies received docs against your rules, and escalates anyone who goes unresponsive — all while posting you a daily digest so you're never surprised.
It works for accounting (client tax docs), insurance (submissions), healthcare (prior auth), financial services (KYC), HR onboarding, vendor compliance, or any custom document collection workflow.
I need to ask you a few setup questions. This takes about 15–20 minutes and you won't have to redo it."
Ask the following questions. You can ask them together in a single message grouped by topic. Do NOT ask more than two groups at once.
Group A — About you and your workflow
"What's your name and role, and what kind of documents are you chasing? (For example: accountant chasing client tax docs, HR coordinator collecting onboarding paperwork, insurance broker collecting submissions, vendor manager collecting COIs and W-9s.) A sentence or two is fine."
"Which systems do you use? I need to know:
Group B — Cadence, verification, and your voice
"How aggressively should I chase? Pick a cadence:
Standard is the right default for most workflows."
"Should I verify received documents before marking them complete? If yes, describe what 'correct' looks like for each document type you handle. (For example: W-9 must have Name, TIN, signature, and date. COI must show your company as certificate holder and coverage above $2M.) If you'd rather just log receipt and let a human verify, say so."
"Paste 2–3 actual nudges you've sent before — copy them from your sent folder. This is the most important setup step: I'll mirror your exact tone, phrasing, and sign-off. If you don't have samples handy, describe your style in a sentence and I'll draft something for you to edit."
After the user answers, summarize what you heard and confirm before writing config:
"Here's what I'm setting up:
- Use case: [use_case]
- Tracker: [tracker_system] at [tracker_location]
- Doc storage: [doc_storage_system] at [doc_storage_location]
- Email: [email_provider], signed [sender_signature]
- Digest + escalations: [digest_channel], escalation owner [escalation_owner]
- Cadence: [nudge_cadence]
- Verification: [enabled/disabled]
- Voice: [1-line summary of tone/style]
Does that look right, or anything to adjust?"
Wait for confirmation before writing.
Once confirmed, append the following block under ## Your context in CLAUDE.md. Fill in every field from the user's answers. Use the exact YAML format below — do not skip fields.
## Your context
use_case: "[accounting | insurance | healthcare | financial_services | hr_onboarding | vendor_compliance | custom]"
tracker_system: "[Airtable | Notion | Google Sheets | Salesforce | HubSpot | other]"
tracker_location: "[base/page/DB name]"
tracker_schema: |
[Describe the columns as the user provided, or use the default schema below if they didn't specify:
Each row represents one document owed by one counterparty.
Columns: Counterparty (text), Counterparty email (email), Document type (single select),
Status (single select — Outstanding | Received | Verified | Failed | Escalated),
Date requested (date), Last nudge sent (date), Nudges sent count (number),
Doc storage link (URL), Notes (long text)]
doc_storage_system: "[Google Drive | Dropbox | SharePoint | Box | other]"
doc_storage_location: "[folder path or root]"
email_provider: "[Gmail | Outlook]"
sender_signature: "[exact sign-off from user's samples]"
nudge_cadence: "[gentle | standard | aggressive]"
escalation_owner: "[Slack handle]"
verify_received_docs: [true | false]
verification_rules: |
[Rules per doc type as described by the user, or "Not configured — mark as Received on receipt." if disabled]
digest_channel: "[Slack channel or DM]"
voice_samples: |
[Paste the user's actual nudge samples verbatim here]
nudge_content_rules: |
Every nudge must include:
- What's needed (specific list, not "your documents")
- Why we need it (the deadline or consequence)
- How they can provide it (link, email reply, etc.)
- Who to ask if stuck
Never include:
- Passive-aggression ("just following up AGAIN")
- Guilt or blame
- Threats unrelated to actual consequence
- Anything that contradicts the user's voice samples
After writing the config, guide the user to connect each required account through Gamut's account connection flow:
"Now let's connect your accounts. I'll need access to:
- [tracker_system] — to read outstanding rows and update statuses
- [doc_storage_system] — to scan for newly received documents
- [email_provider] — to send nudges and acknowledgements (I'll draft, you can review before anything sends if you prefer)
- Slack — to post your daily digest and tag your escalation owner
Connect each one via the Accounts panel in Gamut. Let me know when they're all connected and I'll run a quick read-only check."
Once accounts are connected, run a dry-run check:
Report back what you found:
"Connected and verified:
- Tracker: [N] rows visible, [N] outstanding
- Doc storage: last file was [filename] on [date]
- Email: authorized to send as [sender]
- Slack: [digest_channel] is reachable
Everything looks good. You're set up."
Close with a suggested first action:
"Your agent runs every weekday at 8:00 AM. To see exactly what it would do before it sends anything, try this prompt:
'Check my tracker and document storage but do NOT send any nudges and do NOT update any rows. Show me what you'd do today — which counterparties would get a nudge, which docs you'd verify, which you'd escalate.'
Once the plan looks right, run it again without the skip — that's day one."
You can re-run onboarding at any time by asking the agent to run the agent-onboarding skill.