agents-loops-goalsUse to make a workspace run itself and to build it iteratively rather than in one pass — translate an outcome goal into a build and measurement loop, create workspace agents with scoped tools, and close the loops (scheduled sources, signal watches, event workflows, the await_rows return edge, reply
build-gtm-workflowUse when building a repeatable GTM workflow (sourcing → enrichment → qualification → copy → enrollment, or any subset) as a workspace TABLE that an agent operates — via Command or Claude Code. Covers the build-time wiring (create table, add module columns, map inputs/outputs, set the cascade), the d
closed-loopUse once a campaign is live and the question becomes "how do I keep it running and improving". The entry point to the closed outbound loop — find, enrich, enroll, handle replies, keep deliverability healthy, optimize, draft the next playbook from what won, and start again — and the rules every loop
deliverabilityUse daily, or whenever replies drop, bounces rise or the user asks "do our mails still land in the inbox". Reads inbox and domain health (SPF, DMARC, DKIM), bounce and reply rates and seed-test placement, sorts every sending inbox into burned / untested / healthy on real data, and proposes the small
draft-gtm-playUse when starting a NEW GTM campaign/play from scratch, or when the user asks "how do I run a campaign here", "draft a play", "set up outreach for X". The methodology layer ABOVE build-gtm-workflow — guides drafting the STRATEGY first (the playbook as a strategic paper: ICP → persona → offer → proof
enroll-outreachUse when qualified leads are waiting and should enter a sequence — review who is ready, prove each lead passes the enrollment gate (suppression, history, rendered copy, pre-send checks), fit the batch to sender capacity, and enroll after an explicit yes. Enrollment is real outbound to real people, s
gtm-handoffsUse when connecting the pieces of a GTM Automation workspace — how Wissen assets, Playbooks, Quellen, Tabellen, Sequenzen and Workflows link to each other, which mechanism and tool creates each connection, how a Workflow actually runs (trigger → steps → dry run → live), and which connections do NOT
gtm-operateUse when an AI agent (Claude Code, Codex, a customer's own agent) needs to OPERATE a GTM Automation workspace over its tools — read tables/Wissen, source and enrich rows, run columns, spend credits safely — via either the `gtm` CLI (shell) or the workspace MCP tools (docked at /api/mcp, key as Beare
gtm-quickstartUse to take a GTM Automation workspace from empty to a live, sending campaign in one working session — the complete build order: intake interview, Wissen assets, playbook, table (source → qualify → enrich → copy → enroll), sequence, workflow, first bounded run, and the measurement loop. Start here f
health-checkUse when the user asks "how is everything doing", wants a status snapshot of the workspace, or before any change to a running campaign. Read-only — reports active playbooks and sequences, what actually went out against capacity, silent failures (runs that say done and delivered nothing), sender and
optimize-outreachUse weekly, or when a campaign under-performs, to decide what to change — measure every sequence, step and A/B variant on real replies and positives, judge each against a bar fixed in advance (keep, stop, scale, unclear), and write at most three concrete proposals, each with its evidence, the exact
outbound-playbookUse when deciding WHAT to build in a GTM Automation workspace, not how — choosing an ICP, picking a channel, setting cadence and volume, writing copy that survives spam filters, and judging whether a campaign is working. The go-to-market method behind the mechanics, with the numbers it rests on (ben
playbook-from-winnersUse when a new offer, audience or channel is about to be opened, or a motion has a clear winner worth repeating — mine what provably converts in this workspace (playbooks, sequences, variants, angles, the replies themselves), pick the closest analogous winner, distil the pattern and the reason it wo
playsUse when the user names a motion to build — local business outreach, LinkedIn post engagers, keyword/social signals, job-opening scraping, inbound lead qualification, website-visitor outreach, lost-deal reactivation, CRM blacklist reconciliation, reply-to-meeting, or mining a CRM for the ICP. Ships
reply-triageUse when the user wants to work the reply inbox or check for new responses — surface inbound replies, read the whole thread, classify each one (positive, question, objection, referral, not interested, opt-out, out of office), draft answers, and hold every send for a per-reply yes. Opt-outs are suppr
sequencesUse when building or changing the touch plan a lead experiences over time — sequence steps and delays, the seven step kinds, slot copy, sender binding and daily limits, enrollment, branching, and the stop rules. Covers email, LinkedIn and WhatsApp. Read copy-patterns.md next to this file for the per
setupUse FIRST, right after installing the gtm plugin, to connect Claude Code to a GTM Automation workspace. Guides the agent through installing the `gtm` CLI, logging in with the workspace MCP key, and validating the connection with `gtm whoami` + `gtm tools`. Once connected, go to gtm-quickstart for th
sourcing-topupUse when a playbook or sequence is running low on qualified leads, or the user wants to refill the pipeline — check supply against what tomorrow's sends need, then top up sourcing and run enrichment and qualification within a hard credit cap. Reads and dry runs happen immediately; starting a source,
validate-gtm-thesesUse BEFORE drafting a play or building a table, whenever a new motion, vertical or audience is being opened — and whenever the user says "test what works", "which hypothesis", "A/B", "validate", "we're not sure this is the right audience". The methodology layer ABOVE draft-gtm-play: a campaign is on