Imported from tiga-labs/tiga-gtm (skills/sequence-builder/SKILL.md). Install upstream with npx skills add tiga-labs/tiga-gtm --skill sequence-builder. Copyright stays with the author.

Sequence Builder Skill

Author sequence steps and their AI personalizations (p13ns) — the content side of outbound sequences. This is the authoring half of the sequence lifecycle; the running half — enrolling and removing people, activating, and monitoring performance — is the sequence-runner skill. If the task is about who's in the sequence or how it's performing rather than what the steps say, hand it to sequence-runner.

Read before starting:

• references/api-reference.md (Sequences API and Sequence Steps API)
• tiga-gtm/docs/merge-fields.md — important: email_body and linkedin_message are HTML fields; merge fields in them must use the <span class="tiga-merge"> format, not plain {{.FieldName}} syntax
• tiga-gtm/skills/sequence-builder/references/p13n-api.md for full p13n API detail and prompt-writing guidance

Key Concepts

Sequence must be inactive to modify steps. You cannot add, edit, or delete steps while a sequence is enabled. Always deactivate first, make changes, then reactivate.

Email signature is appended automatically. Tiga appends the sending user's stored email signature to every outgoing email at send time. Never write a signature or the sender's name at the end of an email_body — end the email with a short closing like Best, and nothing after it. Do not use {{.UserName}} as a sign-off. (Emails only — LinkedIn messages do not get a signature appended.)

Field casing differs by endpoint:

• Add-step requests use semantic snake_case fields such as step_name and step_instructions.
• PATCH requests accept snake_case or PascalCase aliases. Use snake_case consistently, always include action, and send only fields relevant to that action. SequenceEmail updates must include both email_subject and email_body.
• GET and create responses use the persisted step model and are mostly PascalCase. Do not copy an entire GET response into a PATCH request or mix its PascalCase fields with snake_case fields.

Step action types:

P13n support: SequenceEmail and LinkedInMessage support AI personalization via merge fields. UserTask and PhoneCall do not — they show instructions to the rep, not outreach content.

Human approval: the add-step field requires_human_approval works on every step type. On SequenceEmail, LinkedInMessage, and LinkedInConnect steps, true holds each generated message in the approval queue — nothing is sent until a rep approves it. On UserTask/PhoneCall steps, true means the rep must mark the task done before the sequence advances. If the user wants to review emails before they send, set this flag on the email step itself. ALWAYS set the first step of a sequence requires_human_approval = true. This ensures that the user will be able to review any sequence before sending.

What is a p13n? An AI-generated text snippet written fresh for each person the moment their outreach task is created. It is step-linked (attached via step_id), ephemeral (recomputed per person at task creation, not stored as a standing fact), and template-embedded (referenced in the step's content as {{.key}}).

P13Ns vs. Signals:

Use a p13n when the output appears directly in email or LinkedIn content. Use a signal (signals skill) to store a fact for filtering, scoring, or routing.

How p13ns connect to steps:

1. Create a step (returns the step's ID)
2. Create a p13n with step_id = <step ID> (returns a key like personalized_opening_a1b2c3)
3. Update the step's email_body or linkedin_message to include {{.personalized_opening_a1b2c3}}
4. When a person is added to the sequence, the system auto-computes the p13n and substitutes the merge field

Workflow A: Create a New Personalized Step

Two ways to personalize a step — pick based on whether the personalization already exists as data you can put in a CSV, or needs to be AI-generated per person:

• You already have (or can build) a CSV of people with the personalization as a column — e.g. a pre-written intro sentence, a scraped fact, anything you can compute or already know per person outside of Tiga. Use references/p13n-with-custom-columns.md. No p13n/AI generation involved — the value is just imported as a custom column and referenced as a plain {{.field}} merge field.
• You don't already know the list of people, or the personalization needs to be generated by AI per person at task-creation time (e.g. "write a sentence referencing their recent LinkedIn activity"). Use references/p13n-with-ai-custom-section.md — this is the p13n workflow described above.

Workflow B: Update Existing Step Content

Use when: The step already exists and you want to change its email body, subject, or LinkedIn message.

1. Get sequence description to find the step ID:

   curl -X GET "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/description" \
     -H "X-Tiga-Auth: $TIGA_API_KEY"

2. Deactivate the sequence:

   curl -X POST "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/deactivate" \
     -H "X-Tiga-Auth: $TIGA_API_KEY"

3. Read the current step content before patching — required for SequenceEmail steps where you need to preserve one of the two fields (subject or body):

   curl -X GET "https://app.tigalabs.com/api/v1/step/<step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY"

4. Update the step:

   For email steps (email_body is HTML — use <span class="tiga-merge"> for merge fields):

   curl -X PATCH "https://app.tigalabs.com/api/v1/step/<step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "SequenceEmail",
       "email_subject": "New subject line",
       "email_body": "<p>Hi <span class=\"tiga-merge\" data-custom-column-id=\"null\" data-computed-config-type=\"null\" data-value=\"{{.FirstName}}\" data-entity=\"Person\" data-alt-text=\"\">First Name</span>, <span class=\"tiga-merge\" data-custom-column-id=\"<p13n-id>\" data-computed-config-type=\"<p13n-type>\" data-value=\"{{.my_p13n_key}}\" data-entity=\"AiSection\" data-alt-text=\"\">My P13n Label</span></p>"
     }'

   For LinkedIn steps (linkedin_message is also HTML — same span format):

   curl -X PATCH "https://app.tigalabs.com/api/v1/step/<step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "LinkedInMessage",
       "linkedin_message": "<p>Hi <span class=\"tiga-merge\" data-custom-column-id=\"null\" data-computed-config-type=\"null\" data-value=\"{{.FirstName}}\" data-entity=\"Person\" data-alt-text=\"\">First Name</span>, <span class=\"tiga-merge\" data-custom-column-id=\"<p13n-id>\" data-computed-config-type=\"<p13n-type>\" data-value=\"{{.my_p13n_key}}\" data-entity=\"AiSection\" data-alt-text=\"\">My P13n Label</span></p>"
     }'

   For task or call steps (instructions is plain text — use {{.FieldName}} directly):

   curl -X PATCH "https://app.tigalabs.com/api/v1/step/<step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "UserTask",
       "instructions": "Call {{.FirstName}} at {{.Phone}} to follow up on the email. Reference their role as {{.Title}} at {{.AccountName}}."
     }'

   curl -X PATCH "https://app.tigalabs.com/api/v1/step/<step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "PhoneCall",
       "instructions": "Call {{.FirstName}} — mention the email you sent and ask about their current outreach process."
     }'

5. Reactivate:

   curl -X POST "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/activate" \
     -H "X-Tiga-Auth: $TIGA_API_KEY"

Note: When you update a step's content, any p13n {{.key}} merge fields that no longer appear in the email body, email subject, LinkedIn message, or instructions are automatically deleted.

Workflow C: Add a Phone Call or Task Step

Use when: Adding a manual reminder step — a phone call prompt or a custom task for the rep. These are simpler than email/LinkedIn steps: no email body, no p13n.

1. Deactivate the sequence (Steps 1–3 from Workflow A)

2. Add the step:

   Phone call step:

   curl -X POST "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/add-step?stepToAppendToId=<last-step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "PhoneCall",
       "step_name": "Follow-up Call",
       "step_instructions": "Call {{.FirstName}} at {{.Phone}}. Reference the email you sent 3 days ago. Ask about their current outreach process.",
       "requires_human_approval": true,
       "can_run_on_weekends": false
     }'

   Custom task step:

   curl -X POST "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/add-step?stepToAppendToId=<last-step-id>" \
     -H "X-Tiga-Auth: $TIGA_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "action": "UserTask",
       "step_name": "Research Before Call",
       "step_instructions": "Review {{.AccountName}}'s LinkedIn page and note any recent company news. Look for a relevant hook for the call.",
       "requires_human_approval": true,
       "can_run_on_weekends": false
     }'

   requires_human_approval: true means the rep must mark the task done before the sequence advances to the next step. Set to false if the step should auto-complete after the delay (see Key Concepts).

3. Activate the sequence:

   curl -X POST "https://app.tigalabs.com/api/v1/sequence/<sequence-id>/activate" \
     -H "X-Tiga-Auth: $TIGA_API_KEY"

Workflow D: Manage P13Ns Directly

List, inspect, update, or delete personalizations independently of step authoring:

• List all: GET /api/v1/p13ns (or ?step_id=<step-uuid> for one step's p13ns)
• Get one: GET /api/v1/p13n/<p13n-id>
• Update: PUT /api/v1/p13n/<p13n-id> — writable fields: label, prompt, word_limit, default_value, temperature
• Delete: DELETE /api/v1/p13n/<p13n-id>

Warning: If a deleted p13n is linked to a step, also remove its {{.key}} merge field from the step's content, otherwise the merge field renders as blank text.

Full payloads, preview-run polling, and prompt-writing guidance with examples: references/p13n-api.md.

Merge Fields

See tiga-gtm/docs/merge-fields.md for the full reference including entity types, display labels, and the common field list (Person, Account, LinkedIn, User, Dates).

Key rule: email_body and linkedin_message are HTML — merge fields in them require the <span class="tiga-merge"> HTML format. instructions (UserTask/PhoneCall) and AI prompt fields (p13n prompts, signal prompts) use plain {{.FieldName}} syntax.

Signature rule: never end an email_body with {{.UserName}} — the user's email signature is appended automatically at send time. Close with Best, and nothing after it.

Get the full list including custom columns: GET /api/v1/mergefields

Related skills: sequence-runner to enroll people in the sequence and monitor performance; signals for persistent facts used in filtering/scoring; flow-builder for play_type: flow agent automations.