--- name: pipeline-check description: Answer where the user's LinkedIn outreach stands in PoliteReach. Use when the user asks how today went, how campaigns are doing, which of a list of people are already in which campaigns, who accepted and what goes out to them next, who is being followed up and when, how many invites are pending, how results, outcomes or an A/B test look, or wants someone added to their do-not-contact list. --- # Where things stand All reads below change nothing unless a step says otherwise. Most are instant; `li_pending_invites` opens the browser. Reads without an account cover every account; scope with `sessionLabel` when the user means one. Ask for the compact form first and widen only when needed: filters and `fields` on `li_list_campaigns`, `view: "summary"` or the default compact rows on `li_campaign_contacts`, `format: "csv"` for a long table. Independent reads (several campaigns, several accounts) can run in parallel. If a tool says PoliteReach could not reach its database, retry once, then report that account as unreadable, never as empty. ## How did today go? `li_daily_digest` per account: invites today and this week, acceptances, messages out, replies in, who is waiting, pending backlog, failures. A blocked account leads the report. ## How are my campaigns doing? `li_list_campaigns`: narrow with `status`, `namePrefix`, `sessionLabel` or `minContacts`; read `truncated`, and while it is true call again with `offset` = `nextOffset`. A campaign not launched yet sends nothing: `status: "not_launched"` lists those. Then per campaign `li_campaign_contacts` with `view: "summary"` (counts over every contact) or `li_campaign_status`. Report the `byConnectionStatus` and `byConversationState` counts as they are; there is no separate stopped count. `li_campaign_contacts` rows are paged: read `truncated` before calling a page the whole campaign. Rows are compact by default (`stepCounts`, no message text); pass `includeSteps: true` only to read the copy or get step ids. To list only some people, filter on the server: `connectionStatus` (e.g. `["accepted"]`), `conversationState` (e.g. `["needs_reply"]`) or `hasInbound: true`; `matchingContacts` is the count. ## Which of these people are already in a campaign? `li_check_targets` with the `profileUrls` (up to 1000) and `view: "people"`: one row per person listing every campaign they are in, on every account the user can see (connection, message, outcome, inbound), each campaign named once in `campaigns`. Anyone not in the table is in none. When the user is building new campaigns, pass `campaignNamePrefix` (or `campaignId`): anyone already in one of them is named in `batch.members`. The default `summary` view shows only each person's furthest bucket plus `otherCampaigns`. ## Who accepted, and what goes out next? `li_accepted_pending_message`: rows already carry `firstMessageText` and `firstStepScheduledAt`. Flag rows whose `firstStepScheduledAt` is in the past, and any with `sendFailed`. If something no longer fits: - Change the words: `li_edit_step` (reversible). - Hold one message: `li_cancel_steps` (cannot be undone). - Stop the person entirely: `li_mark_stopped` (ends their sequence). Each only on the user's yes. ## Who is being followed up, and by what? Every open thread has one owner: the booking follow-up for people who agreed to a call (`li_bookings` with `status: "agreed"`; `li_booking_due` for who is due), a follow-up chain for any other open thread waiting on them, the campaign sequence for people who never replied, a snooze for "not now". A final verdict, do-not-contact or a flagged thread has none. `li_list_chains` (default `active`, all accounts, or `profileUrl` / `sessionLabel`; read `truncated` and raise `limit`, max 200, until it is false): report each chain on one line: name, account, each step's due date (and `send` or `draft`). Rows in `li_replies_to_answer` and `li_list_conversations` carry the same `activeChain` (null = none); `activeChainUnreadable` / `activeChainLookupFailed` means the read failed, so ask `li_list_chains`, never report "no chain". To add or change one, use the `answer-replies` or `keep-warm-leads` skill. ## How many invites am I waiting on? `li_pending_invites` per account opens the browser and can take minutes; pass `maxInvites` for a quick look. Use `pendingOnLinkedIn` for the count. If `ageDays` is 0 on every row, report no ages. Read `reach`: `partial` or `failed` is not an empty list. Withdraw nothing unless the user asks: a withdrawn invite blocks re-inviting that person for about three weeks. On a clear yes: `li_withdraw_invites` with `confirmWithdraw: true`, then `li_withdraw_status`. `notFound` means gone only when the whole list was read (`reachedEnd: true`); `notReached`/`notReachedCount` are still pending (the walk stopped first, `limitedBy`): never call them gone or re-invite. Invite cleanup follow-up: read `autoWithdraw.lastRun` in `li_list_accounts`: `withdrawn`, `scanned` of `linkedinPendingTotal`, and `notReachedCount` (stale invites not reached yet). Report `invites.backlogWarning` (`li_account_activity`) when set: a warning only, new invites still go out; if `autoWithdraw.enabled` is false, suggest turning on `li_set_auto_withdraw`; if it is on but `autoWithdraw.untracked` is false, say it withdraws only invites PoliteReach sent. Report the account's learned weekly cap (`invitePace.learnedWeeklyCap`) when set. ## How are results? - `li_outcome_summary`: the human verdicts (won, meeting booked, interested, not interested, no outcome). `undecided` means nobody has judged those yet, not that they went badly. Report `repliedNoVerdict` (they replied, no verdict: work owed), not the whole `undecided`; `neverReplied` and `notMessagedYet` have nothing to judge. Suggest the `answer-replies` skill to close them out. - `li_outreach_analytics` compares acceptance and reply rates by strategy, ICP or campaign. For reply rates quote `repliedEverRate` (how many people replied), never `replyRate` (only replies still awaiting a verdict). To compare many campaigns: `groupBy: "campaign"` with `namePrefix` or `status`, `fields: ["contacts", "acceptanceRate", "repliedEverRate"]`, or `format: "csv"`; page with `nextOffset`. - An A/B test made with `li_experiment_assign`: `li_outreach_analytics` with `groupBy: "arm"` and `experimentId` (without it, it lists the experiments). Each rate (acceptance, reply, booking, `positiveReply`) carries `k`, `n` and a 95% interval: quote `positiveReply` with its `n` and interval, and say when the intervals overlap, which means no winner yet. `forcedToControl` people (no post) are outside the comparison. `acceptanceByEngagement` (reacted / notReacted, `unclear` in neither) explains an arm's acceptance; it is not randomised, so compare arms first and never call it the effect of reacting. - Invite pace: from `li_list_accounts` (`invitePace`). It is PoliteReach's own per-tier safety policy, not a LinkedIn rule, and cannot be raised. Never quote one flat number for everyone. ## Never contact these people `li_suppress_profiles` with the `profileUrls` and a `reason`: preview with `confirmSuppress: false`, then `confirmSuppress: true`. Listed people are refused at import, before every invite and before every automated message. Read the list with `li_list_suppressed`; remove one with `li_unsuppress_profile`. ## What is the account doing right now? `li_account_activity`: what runs now, what is queued, and why anything waits. "Queued" usually means another job is ahead on a shared browser. For setup or sign-in problems, use the `get-started` skill.