clickmax-playbooks

Use when the user wants to create, edit, delete, link to pipelines, preview, or rehearse a CRM playbook (the sales method with stages, questions, tones, keyword triggers and guardrails that guides mee

SKILL.md

GitHub

When this applies

Use for the sales-method document in Contatos & Comercial › Configurações › Playbooks (Contacts & Sales › Settings › Playbooks): write or change one, make pipelines suggest it, preview what a meeting room receives, rehearse it with the AI copilot, or delete it.

Not this skill:

  • pipeline board, stages, cards, pipeline settings -> clickmax-pipelines
  • lead search / identity -> clickmax-leads
  • an AI agent's own persona/knowledge -> not exposed here; this skill only picks an existing agent by id

Key assumptions

  • A playbook is an independent entity, not a pipeline setting. Pipelines only SUGGEST one when a meeting is booked; whoever books can pick another playbook or none. Suggestion applies only when the booking does not choose (public booking pages, automations, AI bookings, or a scheduled_calls_create that omits playbookId).
  • Effect follows the scopes: meeting is the Click Meeting room panel (Roteiro column + AI copilot briefing) and the voice-call window; livechat is the playbook block of the LiveChat lead cockpit (progress kept per conversation, playbooks_list with scope: "livechat" lists the eligible ones). whatsapp and flows are shown as "coming soon" and are rejected on save.
  • Structure: stages (title, objective, talk track, watch-out, questions = the meeting checklist) · tones (name, when, delivery) · keyword triggers ("deixas") · attachments (existing http(s) URLs) · guardrails (what the copilot must never suggest). Limits: 20 stages, 30 questions/stage, 12 tones, 40 triggers, 12 terms/trigger, 30 attachments, 20 guardrails.
  • Save is WHOLE-DOCUMENT (playbooks_update = PUT). Omitted collections are wiped, attachments included, and they cannot be re-uploaded through this surface. Always read-modify-write from playbooks_get.
  • Ids (stage, question, tone, trigger) are authored by the caller (1-40 chars). Questions must be unique across the WHOLE playbook, not per stage: the room marks a question as covered by id. Keep existing ids stable on edits; a changed id is a different item.
  • Keyword triggers are deterministic and model-free: the room compares each GUEST utterance with the terms (case, accents and punctuation ignored, whole words only: caro fires on "ficou caro demais", not on "carro alugado"), shows the same cue text instantly, and the same trigger returns at most every 5 minutes. Only the guest speaking fires them. Terms need at least 2 chars.
  • Editing a playbook changes meetings booked AFTER the save and future calls. A booked meeting keeps the script it received; the host refreshes it by re-picking the playbook inside the room.
  • Delete is hard and permanent. Pipelines that suggested it are left with no suggestion; booked meetings keep their room script.
  • Writes (create/update/delete/link/simulate) are for owner/admin/editor; every role can read.
  • The talk track accepts 6000 chars but the meeting panel shows the first 2000. playbooks_meeting_guide_get shows the truncated, real version.
  • Not available here: audio transcription for the simulator, file upload for attachments, AI-generated playbooks ("Com IA" in the editor), templates gallery (NEPQ, SPIN, BANT, Challenger, consultative diagnosis are editor templates, not API objects).

Thought process

  1. Find out what the user has: playbooks_list first; edit before creating a near-duplicate.
  2. New method from scratch: interview briefly (sales motion, buyer, stages the conversation goes through, top objections, forbidden promises), then design; do not invent pricing or promises the user never gave.
  3. Separate three jobs: authoring the document, linking it (suggestion per pipeline), choosing it for one meeting.
  4. Validate a draft before or right after saving with the rehearsal; treat its output as one sample, not a verdict.

Execute guide

  • Build: draft stages (one clear objective each, 3-8 open questions that steer the buyer, a short talk track, a concrete watch-out), 2-5 tones for moments such as opening/objection/closing, keywords for predictable moments (price, competitor, "preciso pensar") and guardrails for hard "never" rules. Then mcp__plugin_clickmax_clickmax__playbooks_create with scopes: ["meeting"] and [] for collections you have nothing for. Generate ids like stage-1, q-1-1, tone-1, cue-1.
  • Edit: mcp__plugin_clickmax_clickmax__playbooks_get -> change only the intended items -> mcp__plugin_clickmax_clickmax__playbooks_update with the full document, ids preserved.
  • Make pipelines follow it: mcp__plugin_clickmax_clickmax__playbooks_pipelines_list shows every pipeline with linked and currentPlaybookName. Attaching replaces the pipeline's current suggestion, so confirm when currentPlaybookName is not null. Then mcp__plugin_clickmax_clickmax__playbooks_pipeline_attach / mcp__plugin_clickmax_clickmax__playbooks_pipeline_detach. From the pipeline side use mcp__plugin_clickmax_clickmax__pipelines_playbook_get (returns the playbook and the default copilot agent) and mcp__plugin_clickmax_clickmax__pipelines_playbook_set: absent field = untouched, null = clear; defaultCopilotAgentId is a separate concern from the method and no tool here lists AI agents, so only set it with an id the user gives.
  • One meeting: mcp__plugin_clickmax_clickmax__playbooks_meeting_guides_list gives the choices (ref is the playbook id); pass it as playbookId on mcp__plugin_clickmax_clickmax__scheduled_calls_create, or change it with mcp__plugin_clickmax_clickmax__scheduled_calls_update (null = no playbook, omitted = keep, and creating without it falls back to the pipeline suggestion).
  • Preview: mcp__plugin_clickmax_clickmax__playbooks_meeting_guide_get returns what the room receives (panel guide, narrated script, guardrails).
  • Rehearse: mcp__plugin_clickmax_clickmax__playbooks_simulate runs the real copilot on a made-up conversation. It has no playbook id: pass the method through briefing. For a saved playbook take script and guardrails from playbooks_meeting_guide_get; add briefing.goal and briefing.persona. Send the whole turns list each call (host/guest); a guest line as the last turn mimics how the room wakes the copilot. Vary reason (objection, question, silence...) to probe stages. Use 3-6 scenarios at most (each is a real AI call).
  • Trigger check: the simulator does not run keyword triggers. To verify one, compare the sample sentence with the trigger terms by hand using the matching rule above and say so.
  • Delete: mcp__plugin_clickmax_clickmax__playbooks_list -> read pipelinesCount -> state how many pipelines lose their suggestion -> mcp__plugin_clickmax_clickmax__playbooks_delete only after explicit confirmation.

Report

  • Lead with the job: created / edited / linked / previewed / rehearsed / deleted, then the playbook name.
  • Created or edited: stages count, questions count, tones, triggers, guardrails, scopes; on edits list what changed and what was kept.
  • Link changes: pipelines added/removed by name, plus any suggestion that was replaced.
  • Rehearsal: per scenario show the last guest line, then the suggestion (title, text, tone) or "no suggestion" with reason; separate "the copilot chose silence" from "the AI call failed" (reason starting with ai_action_). End with what to tweak in the playbook, offered as opt-in.
  • Cap long lists with +N more; follow-up mutations are opt-in.

Warnings

  • playbooks_update without the full current collections silently deletes them; never send a partial document.
  • Attaching to a pipeline that already suggests another playbook replaces it without asking the pipeline's owner.
  • Deleting a playbook cannot be undone.
  • Rehearsal answers vary between calls and cost AI calls; one silent answer is normal (most turns need no card) and does not mean the playbook is broken.
  • Do not promise that editing fixes meetings already booked.

Anti-patterns

  • Treating the playbook as a pipeline setting, or trying to create one per lead.
  • Regenerating ids on every edit.
  • Writing prices, discounts or deadlines into stages or triggers that the user never provided.
  • Marking whatsapp/flows scopes.
  • Looping the simulator to "optimize" the text without the user asking.
  • Setting the pipeline's default copilot agent as a side effect of linking a playbook.

Clickmax skill revision: 30bc1243c2ed