Build a closed-won lookalike outbound agent from CRM deal context

Start by giving Claude Cowork the reference material it will use on every run instead of letting it infer your market from one closed-won deal. Create four Markdown files or Google Docs named ICP, Personas, Messaging, and Product Facts, then include positive-fit traits, disqualifiers, buying committee roles, pain language, proof points, competitor context, security/compliance caveats, and claims to avoid. Open Claude Cowork, paste the prompt below, attach or reference the draft files, and ask it to turn them into a clean reusable skill pack with a short test section at the end. Save the final skill files in the shared folder Cowork can access and log the file URLs in Google Sheets. QA check: ask Cowork to score three known good-fit accounts and three bad-fit accounts before you let it read live CRM data. Prompt handoff: open Claude Cowork, paste the prompt template below, attach or reference the inputs named in the prompt, and copy the output into Google Sheets before continuing.

Open Claude Cowork and paste this prompt. Use the draft ICP, persona, messaging, and product notes as inputs, then save the cleaned skill files to {{skill_file_folder}} and log the final links in {{run_log_location}}.

Create a reusable foundation skill pack for a closed-won lookalike outbound agent.

Draft ICP notes:
{{icp_notes}}

Draft persona notes:
{{persona_notes}}

Draft messaging notes:
{{messaging_notes}}

Product facts and approved claims:
{{product_facts}}

Claims, audiences, or account types to avoid:
{{claims_and_accounts_to_avoid}}

Return:
1. ICP skill file
2. Persona skill file
3. Messaging skill file
4. Product facts skill file
5. Disqualifier checklist
6. Example good-fit account scores
7. Example bad-fit account scores
8. Questions that must be answered before the agent can run

Do not invent claims, customer names, metrics, or integrations. Mark unknowns clearly.

Connect Claude Cowork to the smallest set of systems needed for this workflow: HubSpot for closed-won deal context, Clay for enrichment tables, Slack for approvals, and email for draft creation or export. Restrict HubSpot access to deals, companies, contacts, associations, and the call-intelligence fields that describe pain, decision criteria, objections, and products purchased. In Clay, create a dedicated table for this workflow instead of letting the agent write into an active production outbound table. Open Claude Cowork, paste the prompt below, and ask it to produce a permissions map, read/write boundaries, and a test procedure using one harmless closed-won deal. Save the permission map in Google Sheets and do not schedule the agent until the read/write test posts only a summary to Slack. Prompt handoff: open Claude Cowork, paste the prompt template below, attach or reference the inputs named in the prompt, and copy the output into Google Sheets before continuing.

Open Claude Cowork and paste this prompt. Use the system list and access policy below, then save the permission map to {{run_log_location}}.

Create the connection and permission plan for a closed-won lookalike outbound agent.

Systems to connect:
{{systems_to_connect}}

Required HubSpot objects and fields:
{{hubspot_objects_and_fields}}

Clay table name and allowed columns:
{{clay_table_spec}}

Slack review channel:
{{slack_review_channel}}

Email rules:
{{email_rules}}

Return:
1. Required connections
2. Read-only permissions
3. Write permissions
4. Actions the agent is not allowed to take
5. One-deal test procedure
6. Expected Slack test message
7. Rollback plan if the test writes to the wrong place

Do not create live outbound tasks or send emails.