Aldchiw/prima-gtm-skills

13 skill entries

• impeccable

  Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.

  Needs a keyActs undeclared1,313 words
• impeccable

  Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.

  Needs a keyActs undeclared1,313 words
• prima-committee

  Identifies the buying committee — real, named contacts per account — for accounts already classified `4B`/`4C` (via `sub_segment`) or `Cat1`/`Cat2`/`Cat3` (via `company_category`) by `prima-icp-check` (`4A`/`4D` sub-segments not supported yet). `Cat1` (AI Infrastructure Operators) and `Cat2` (Crypto Miners Pivoting to AI) share a single target-title dictionary — Notion documents the same direct-equipment-buyer profile for both — with one PRINCIPAL tier (VP Infrastructure, Head of Deployment, Director of Supply Chain, VP Operations, Head of HPC, Procurement Director) and a COO FALLBACK gated by the same founder/C-suite-always-FALLBACK rule as `4C`. Sources widely (3-4 verifiable names per account — the "banca") but writes narrowly: the outreach guardrail still caps actual contact at 2 people per account regardless of segment, with the rest held on the bench for later promotion if a contact doesn't respond. Cat3 also enforces a stricter manufacturing-division-only rule and, since its targeting logic is an unvalidated v1 hypothesis (Aldahir, 2026-08-03), sourcing runs on every Cat3 priority but `P1` contacts always come back as `NEEDS_HUMAN_REVIEW` (never auto-cleared `CONTACT_FOUND`) until the hypothesis is validated on `P2` data. Writes/updates an accumulative `output/account_roster.csv` (never regenerated) tracking every sourced name — skill-owned columns vs. operator-owned lifecycle columns (`contact_status`, touches, outcome) — and only hands a contact to `prima-email-waterfall` when it's both `contact_status = active` AND `status = CONTACT_FOUND` — bench contacts and unconfirmed `NEEDS_HUMAN_REVIEW` contacts (e.g. any Cat3 `P1` row, active or not) stay `email_status = NOT_ATTEMPTED` so Deepline credits are never spent on an unpromoted or unconfirmed name. Sources names via free WebSearch (company site + public LinkedIn snippets) first; only falls back to paid Deepline providers (ContactOut, Lusha, RocketReach, etc.) after stopping to show how many accounts need paid lookup and the estimated cost, and getting explicit approval — same gate as the Clay-vs-Terminal pilot. Never fabricates a name — if no verifiable contact is found for a role, outputs the target role/title only. Use this after `prima-icp-check` (and ideally after `prima-signal-scan` has confirmed the target-title door exists) and before `prima-email-waterfall` needs a named person to find an email for.

  Self-containedActs undeclared6,950 words
• prima-draft

  Generates 3 email drafts (E1/E2/E3) per contact — contacts from `prima-committee`, opening hooks from `prima-hook`, verified emails (when available) from `prima-email-waterfall` — by selecting a template from a versioned, file-based template library and filling it with the contact's real, verified hook facts. Under Prima's guardrails — 6-8 lines, signal first, "reliable supply" narrative (not cheap pricing), 10,000 tons/month capacity claim, zero customer name-drops, zero unvalidated certifications (only AISC/AWS), zero AI-sounding language. Never invents a fact to fill a gap; if fewer than 3 hooks exist, reuses the strongest one across drafts with a different angle instead of fabricating a second or third. Always emits `draft_status: BORRADOR` and `audit_status: NOT_AUDITED` as clean metadata — never a marker inside the email text itself — and only `prima-guardrail-audit` may change `audit_status`. Use this after `prima-committee`, `prima-hook`, and (ideally) `prima-email-waterfall` have run; it's the last step before `prima-guardrail-audit`.

  Self-containedInstructions only1,127 words
• prima-email-waterfall

  Given named contacts (the output of `prima-committee`), waterfalls across Deepline email-finding providers to find and verify a corporate email — falls through to the next provider on failure until a verified hit, or leaves the field empty if none succeed. Never returns an unverified guess as if it were good data, and never accepts a personal email (gmail, etc.) as the sendable result. Single contacts or a 10-20 sample run freely; a full batch stops first to show contact count, max provider depth, and estimated cost, and waits for explicit approval. Use this after `prima-committee` has named the buying-committee contacts and before `prima-hook`/`prima-draft` need a real address to send to.

  Self-containedInstructions only908 words
• prima-generate-leads

  The one-line "director de orquesta" for the Data Centers lead-gen pipeline — a salesperson types only a count and a vertical (e.g. "genera 20 leads de Power & Electrical") and this skill runs discovery → prima-icp-check → prima-signal-scan → prima-scope-score → prima-committee → prima-email-waterfall end to end with no follow-up questions, then writes a single clean row per company to leads_final.csv. Only supports Power & Electrical Distribution and Energy Storage (Aldahir's verticals — prima-scope-score's product table isn't portable to Manu's). Stops before prima-hook/prima-draft/prima-guardrail-audit — this produces a qualified, contactable lead list, not outbound copy. Use this whenever someone asks for N new leads/accounts/empresas in one of these two verticals without specifying a company list themselves.

  Self-containedActs undeclared6,138 words
• prima-guardrail-audit

  Final auditor for Data Centers outbound emails — checks a draft (or an E1/E2/E3 batch from `prima-draft`) against Prima's full message guardrail list and returns PASS/FAIL per email, with BLOCKERS and WARNINGS cited line-by-line. Runs unconditionally before any email goes out — no exceptions. Pure auditor: never rewrites or suggests corrected copy, that's `prima-draft`'s job. Use this whenever a draft (or batch of drafts) needs to be checked before sending, even if the user doesn't say "audit" explicitly, e.g. "revisa este draft", "¿esto pasa guardrails?", "prepara esto para enviar".

  Self-containedInstructions only972 words
• prima-hook

  Given an account's already-verified signals from `prima-signal-scan`, extracts up to 3 ranked, specific, sourced (URL-backed) facts to open the email with — the first-line hook. Signal-first by design — never searches for a new signal of its own; only reads deeper into the source URL signal-scan already verified to pull out the exact figure/date/name that makes the opener specific. Never invents a claim. Returns raw verified facts + source, not email prose — `prima-draft` owns tone/guardrails and writes the actual line. If signal-scan came back NO_SIGNAL on everything, returns an explicit `NO_HOOK` status rather than inventing one. Use this after `prima-signal-scan` and before `prima-draft` needs opening material.

  Self-containedInstructions only901 words
• prima-icp-check

  Classifies Data Centers accounts (Power & Electrical, Energy Storage) against Prima's ICP — sub-segment (4A/4B/4C/4D for Cat4; N/A-Cat1/N/A-Cat2/N/A-Cat3/N/A-Cat5 for the other in-scope categories, or G1/G2/G3 for the adjacent-groups overlay), priority (P1/P2/P3, or read from Notion's curated Prima Priority for Cat1/2/3/5), vertical owner (Aldahir vs Manu vs UNKNOWN), an explicit `company_type`, and exclusions. Category 1 (AI Infrastructure Operators) and Category 2 (Crypto Miners Pivoting to AI) are in scope alongside Cat3/Cat4, gated on Notion's "Equipment Procurement Scope" — its own documented "core qualification gate" for those two categories, distinct from Cat4's Disqualify criteria. Category 5 (Integrated DC Campus Builders) is also in scope, gated on Notion's own "Full-Scope Accountability" criterion (Full / Strong / Partial / Disqualify → Cat7); Category 6 (AI Campus Builders / Developers) is out of scope and only used as a signal to find the tenant (Cat1) or the design-builder (Cat5), and Category 7 (GCs & Integrators) is out of scope for now. This is **GATE #0** for the whole pipeline: it runs before any paid enrichment tool, anywhere downstream, and nothing `excluded = yes` or `company_type = unknown`/`wrong-industry` may proceed to a paid step. Use this whenever the user hands over a list or CSV of company domains/account names for the Data Centers GTM channel and wants them qualified, classified, scored, or filtered — even if they don't say "ICP" explicitly, e.g. "revisa esta lista de cuentas", "clasifica este CSV de dominios", "cuáles de estas empresas nos sirven". This is stage 1 of the 7-skill Data Centers pipeline (icp-check → signal-scan → committee → email-waterfall → hook → draft → guardrail-audit) — always run it first on a new account batch, before signal scanning, committee-building, or drafting.

  Self-containedActs undeclared6,443 words
• prima-reference-leads

  Procesa una lista de empresas solicitadas por un usuario (Gaby, Aldahir, Gus, etc.) que no entran al ICP formal pero se quieren contactar (eventos, listas, fotos, pedidos puntuales). Salta prima-icp-check y prima-scope-score; corre signal-scan → committee → email-waterfall y escribe company_category = Reference en leads_master.csv con el esquema real del master, respaldo previo y OK explícito. Úsalo cuando alguien pida leads de empresas específicas por nombre, no por vertical o categoría ICP.

  Self-containedActs undeclared1,440 words
• prima-scope-score

  Given an account already classified by `prima-icp-check` and scanned by `prima-signal-scan`, estimates the probability that it actually subcontracts structural fabrication (steel/enclosures) rather than manufacturing 100% in-house — Notion's "Equipment Procurement Scope" criterion, which almost never has hard public data behind it. Outputs a 0-100 score + tier (`tier_1`/`tier_2`/`tier_3`), always with a traceable rationale citing the specific signals behind it — never a bare number. Two hard overrides dominate the weighted score: explicit "100% in-house/vertically integrated" evidence forces `disqualified_inhouse` regardless of everything else; a confirmed Import Genius import match forces `confirmed_outsources`. Also sets `needs_manual_scope_confirmation` for P1 accounts — a P1 never reaches `prima-draft` on tier alone. Cat4-only, by inclusion: only `company_category = Cat4` proceeds to scoring — `Cat1`, `Cat2`, `Cat3`, and `UNKNOWN` all return unscored (`scope_tier = N/A-<category>`), since Cat1/Cat2's own scope criterion is polarity-inverted from Cat4's (direct in-house operation is what qualifies them, not what disqualifies them) and Cat3 already has a human-confirmed `notion_scope`. Use this after `prima-signal-scan` (needs its capacity-expansion freshness) and before `prima-committee`.

  Self-containedActs undeclared1,740 words
• prima-signal-radar

  v0.1 — watches a caller-supplied list of already-classified accounts for fresh procurement/supply-chain job-opening signals via predictleads_company_job_openings, and appends verified hits to a unified output/signal_radar_feed.csv. Cat3/Cat4 only in v0.1 — Cat1/Cat2 accounts are logged as "watched via news" (v0.2, not built yet) because the job-category tag this skill filters on was validated 2026-08-10 to miss Cat1-style titles. Never fabricates a signal; a job opening without a source URL is discarded, never invented. Runs on a sample of 3-5 accounts first and confirms cost before a full sweep — same standing rule as every other paid-provider call in this repo. Use this when the user wants recurring/radar-style monitoring of a tracked account list, as opposed to `prima-signal-scan`'s one-time WebSearch scan of a single freshly-classified account.

  Self-containedActs undeclared3,327 words
• prima-signal-scan

  Given a domain, scans 5 sources for verifiable Data Centers signals — plant purchasing/supply chain/procurement job openings, capacity-expansion news, recent funding or grants, target-title presence (Sales Navigator proxy), and Mexico→US customs imports (Import Genius). Target titles for job-opening and target-title checks are read fresh from `prima-committee`'s dictionaries (`4B`/`4C`/`Cat3`) — this skill keeps no title list of its own, and matches by function (not exact string) the same way committee does. Outputs a dated, sourced signal list — empty (NO_SIGNAL) if nothing verifiable turns up, and explicitly NOT_CHECKED if a source couldn't be reached this run. Use this after `prima-icp-check` has classified an account and before `prima-committee`/`prima-hook` need something to work with. Never fabricates a signal.

  Self-containedActs undeclared2,293 words