creatop-gtm/apollo-operator / .claude/skills/apollo-go-live
apollo-go-live
VerifiedThe last mile: enroll a graded list into a built (inactive) sequence, get a human to approve activation, and pull contacts on complaint or bad fit. Use to take an Apollo sequence live safely.
Apollo Go-Live (Launch)
Turn a built sequence and a graded list into a live campaign, safely. This is the bridge everyone skips: the List phase gives you a clean list, the Message phase builds the sequence as an inactive draft, and then nothing happens until someone connects the two and flips the switch. This skill is that step, and it is the one place where a wrong move sends real email to real people, so it is deliberate and human-gated.
When to use
- A sequence exists in Apollo as an inactive draft (Message done).
- A graded, verified list exists as contacts in the account, ideally under one Apollo List label (List done).
- Deliverability preflight passes (Infrastructure): mailboxes warmed, schedule set, sending from a dedicated domain.
If any of those three is missing, go back and finish it. Do not improvise the last mile.
The rule that does not move
The machine never flips the switch. Claude enrolls, summarizes, and recommends. A human runs the activation after confirming the sender, the schedule, and the copy. Every tool in this flow sends or commits real email, so every one of them waits for an explicit human yes.
Process
1. Preflight (Infrastructure, non-negotiable)
Run apollo-deliverability first. Confirm warmup is complete (14 days minimum, 21 optimal), bounce risk is low (100% verified), the sending schedule is business-hours/weekdays/timezone-correct (apollo_emailer_schedules_index), and you are sending from a dedicated domain, not the primary. If deliverability is not clean, stop here. You cannot out-send a reputation problem. If the sending stack does not exist yet, that is a sending-infrastructure job, not a go-live one.
2. Pick the sender mailbox
Call apollo_email_accounts_index. Auto-select the mailbox where default: true, unless the operator names another. Confirm it is a dedicated sending domain that redirects to the main site, never the company's primary domain. For volume, you can pass several mailbox IDs to rotate.
3. Enroll the list (nothing sends yet)
Enroll into the sequence while it is still inactive, so contacts queue but no email goes out until activation. Use apollo_emailer_campaigns_add_contact_ids:
idandemailer_campaign_id: both the sequence id.send_email_from_email_account_id: the mailbox from step 2 (string, or an array to rotate).- Enroll by
contact_ids. Collect the IDs of your graded list withapollo_contacts_search(search the list, take each contact'sid), then pass them ascontact_ids. The tool advertises alabel_namesshortcut to enroll a whole Apollo List by name, but it errors with "Required parameter 'contact_ids' missing", first found in July and re-verified 2026-09-14 on a one-contact list into an inactive sequence. Nothing is enrolled, so do not rely on it. Resolve the label to contact IDs yourself and passcontact_ids. Ask the search for ids only,fields: [{"contacts": ["id"]}]: at 100 contacts a page that is 4 KB instead of the 306 KB full record (measured 2026-11-09). - Leave
sequence_unverified_email: false(the default). This makes Apollo refuse unverified addresses, which enforces the verify-everything rule at the door. Do not flip it to true to "get more in." - Check the opt-out before the first enrollment. Confirm the workspace unsubscribe link is on (a UI setting, not exposed through the API) and that the signature carries a physical address. See
compliance.md. - Only contacts can be enrolled. If your people are search results but not yet contacts, enrich and create them first (List), then enroll.
- Per-contact copy goes in at this call, or not at all from here. If each contact has their own subject and body, pass them as
step_content_overridesnow: they apply only to contacts this call enrolls, and a later re-enrollment skips anyone already in. To change one afterwards, useapollo_emailer_campaigns_update_step_content. Read them back withapollo_emailer_campaigns_step_contentbefore activating. Seeapollo-sequence-builder. - Verify enrollment. Each enrolled contact comes back with the sequence id in its
emailer_campaign_ids. A removal (step 6) sets it back to empty. Confirm rather than trusting the call. To check the whole list afterwards, search it withfields: [{"contacts": ["id", "emailer_campaign_ids"]}].
4. Human review (surface, then wait)
Read the sequence back first with apollo_emailer_campaigns_show, so the summary describes what is actually saved (step count, subjects, and the opening of each body), not what you meant to build. The same response carries sending_diagnostics: before activation its issues should read sequence_inactive and nothing else, with email_steps_without_approved_template at 0. Anything more, such as no_schedule, is a fix to make before asking (verified 2026-11-09). Put the schedule name and timezone it reports into the summary. Then show the operator a plain summary and stop:
Sender:
outbound@dedicated-domain.com(dedicated, warmed 21 days). Sequence: "Q3 Trade Show Managers". Contacts enrolled: 312 (all verified). Schedule: business hours, weekdays, contact timezone. Ready to activate?
5. Activate (human only)
On an explicit yes, activate with apollo_emailer_campaigns_approve (sequence id). This flips active: false to true and starts sending on the schedule. It is irreversible for any email already dispatched. Claude does not run this on its own initiative, and never on "just do it" without the step 4 summary shown first.
Then confirm it is sending. A day in, apollo_emailer_messages_search with the sequence id and display_mode: "metadata_mode" returns the counts the UI shows (scheduled, sent, delivered, bounced, replied). Scheduled but not sent inside the send window is the thing to catch early; read the schedule section below before calling it broken. Nothing scheduled at all, with contacts enrolled, usually means the touches are not approved. Read the sequence with apollo_emailer_campaigns_show and check every touch's status: only approved sends, and to_be_reviewed holds the whole sequence silently. Verified 2026-09-14 on a sequence that sat active for two weeks with five contacts and zero sends.
6. After it is live: pulling contacts
Apollo already auto-handles the normal cases: every sequence built here defaults to finishing a contact on reply or interest and pausing on out-of-office. So you do not manually pull someone just because they replied. Use apollo_emailer_campaigns_remove_or_stop_contact_ids for the explicit cases: a complaint, a removal request, a wrong-person or do-not-contact, or a list you need to yank. Pass contact_ids, emailer_campaign_ids, and mode: "stop" (with a stop_reason) to halt them where they are, or mode: "remove" to take them out entirely.
7. The kill switch: stopping the whole send
Pulling contacts handles individuals. When the problem is the campaign itself, a bounce spike, wrong copy that got through review, or the wrong list enrolled, you need to stop everything at once.
Use the CLI or REST for this, not MCP. The Apollo UI, the CLI, and REST (POST /emailer_campaigns/{sequence_id}/abort) all stop a sequence in one action. MCP can too, through apollo_sequences_update with active: false, but that update is declarative: any step you do not echo back is deleted. Mid-incident is the wrong moment to get that right. The safe MCP version is in apollo-operator.
apollo sequences abort --id <sequence_id> # deactivate, stops sending
apollo sequences archive --id <sequence_id> # retire a sequence you are done with
Install the CLI before you activate anything, not during an incident, because an incident is exactly when you do not want to be running a Homebrew install and an OAuth flow. Setup is in apollo-operator.
Abort first, diagnose second. A paused campaign costs you a day; a bounce spike left running costs you the domain. Then work the incident with apollo-deliverability.
Read the schedule before you activate, because it decides when anything happens
apollo sequences schedules returns more than the id you need for --schedule-id. The useful part is schedule_hash, and it is worth reading before every launch:
{"id":"…","name":"Normal Business Hours","time_zone":"America/Los_Angeles",
"use_contacts_time_zone":true,"skip_holidays":true,
"schedule_hash":{"monday":`8,17`,"tuesday":`8,17`,"wednesday":`8,17`,
"thursday":`8,17`,"friday":`8,17`}}
Three fields change what activation actually does:
schedule_hashgives the real send window per weekday as[start_hour, end_hour]. Check it against the sending rules you are working to. Apollo's stock "Normal Business Hours" is 8 to 17, which is not the same as a 9 to 18 policy, and nobody notices the hour of drift until they look.use_contacts_time_zone: truemeans the window is evaluated in each recipient's timezone, not the schedule's. This is what you want for a US list, and it means activating at one moment produces sends spread across hours. A US-wide list activated at 09:00 Eastern sends to the East Coast immediately and holds the West Coast for another two hours.skip_holidayssilently defers sends on holidays, which will look like a broken campaign if you are watching the queue that day.
So an empty message queue right after activation is usually correct, not broken. Before diagnosing anything, check the local time for the contacts in the list against the window. Only treat it as a fault once a contact is inside their own window and still has nothing queued.
Enrollment fails silently. Check the response body, never the exit code.
Seen 2026-08-31, and it cost a false "done" in a real session.
Five ids were passed as apollo sequences add-contacts --contact-id $IDS in zsh, which does not split an unquoted variable into words the way bash does. So the CLI received one argument, "id1 id2 id3 id4 id5", and Apollo looked up one contact by that id, found nothing, and returned HTTP 200 with exit code 0. (This library first blamed the CLI for joining the ids. That was wrong: the CLI sends --contact-id values as a proper array. The shell joined them.)
{"contacts": [],
"skipped_contact_ids": {"6a95… 6a95… 6a95… 6a95… 6a95…": "contact_not_found"}}
Nobody is enrolled, nothing warns you, and the shell says success. The failure only shows up when a human opens the sequence and sees it empty.
Two rules, both mandatory:
- Assert on the response body after every enrollment.
contactsmust have the length you expect andskipped_contact_idsmust be empty. Treat a non-emptyskipped_contact_idsas a failure even on a 200. The skip reasons are informative (contact_not_found, and others for suppression and ownership rules), so print them. - A workspace admin can keep contacts out, and no call flag gets past it. Apollo's admin settings (Settings, Outbound, Sequences, Enrollment criteria) can require a verified email, a phone number for call steps, a LinkedIn URL for LinkedIn steps, no other active sequence, allowed contact or account stages, and an owner. A sequence ruleset can set its own. They apply to every source: UI, workflows, API, and MCP. Verified 2026-11-09 with "skip contacts active in another sequence" on: a default enrollment was still caught first by the call's own guard (
contacts_active_in_other_campaigns), and an enrollment sent withsequence_active_in_other_campaigns: truewas skipped by the admin rule instead, asenrollment_criteria_active_in_other_sequence, withskipped_contact_detailsnaming the conflicting sequence's id and name. So any reason startingenrollment_criteria_means a workspace rule, not your call: retrying will not help, and changing it is the admin's decision, which matters when you work inside someone else's account. Skipped contacts also appear in that settings page's "View skipped contacts" log. - Pass ids as separate words. On the CLI, write them out (
--contact-id id1 id2 id3), or in zsh expand the variable with${=IDS}. Over REST, sendcontact_idsas a JSON array toPOST /emailer_campaigns/{id}/add_contact_ids. Any of the three works; an unquoted$IDSunder zsh does not.
Enrolling does not send while the sequence is inactive. Enroll first, verify the count, then activate as a separate deliberate step. That ordering gives you a real checkpoint between "the right people are in" and "email is going out."
Common mistakes
- Enrolling into an already-active sequence. Then it sends immediately, no review gate. Enroll into the inactive draft, review, then approve.
- Flipping
sequence_unverified_emailto true. That defeats verification and feeds bounces straight into your domain reputation. - Going live before warmup finishes to save a week. You lose the domain instead of saving the week.
- Sending from the primary domain. One reputation, and it is the company's real one. Always a dedicated sender.
- Treating a reply as something you must manually remove. Apollo finishes them on reply by default. Save the remove/stop tool for complaints and bad-fit pulls.
- Activating without a kill switch installed.
sequences abortis one call on the CLI and REST. The MCP route exists but deletes steps if done carelessly. Get the CLI working before go-live, not in the middle of an incident. - Diagnosing before aborting. Stop the send first. The campaign is still there to investigate once it is no longer making the problem worse.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |