Review — neuro.educarte

All required sections for the principles template are present in system.md: Identity + Mission, Voice, Flow (with STEP 0 keyword check, phases, energy reading, family derivation, NO_RESPONSE), Calibration (energy states), Principles (4 stated), Tools (4 tools listed), and NO_RESPONSE (mandatory section). Content matches template intent.

examples.md contains 14 examples covering all 7 base scenarios (happy path Ej1, resource/link delivery Ej2/2b, objection handling Ej6, disqualification Ej7, emotional/sensitive Ej5, re-engagement Ej14, off-topic Ej11) plus two conditional scenarios: keyword trigger (Ej2, Ej10, Ej12, Ej13) and family/parent objection (Ej5, Ej12). Total count is 14, under the 20-example cap.

Every example in examples.md follows the flow steps defined in system.md (STEP 0 keyword check → fork familia/profesional → conocer al lead → empatía + pitch → booking or NO_RESPONSE). Each example reaches a terminal state (booking link sent, PDF sent, NO_RESPONSE, or derivation to Adri). No example ends mid-flow.

system.md documents explicit exceptions to the universal fork (familia/profesional question). STEP 0 states: 'Excepciones (saltar el fork): (a) Lead ya se identificó... (b) Lead escribe explícitamente... (c) Keyword MODELO con profesión ya conocida... (d) Keywords del taller en vivo... (e) is_professional=true en el contexto'. These are named carve-outs, not contradictions. The default rule and its exceptions coexist under different preconditions.

All SDK files align on behavior and data. system.md's STEP 0 exception for MODELO (skip fork when profession known, go direct to PDF) is consistent with keywords.json's MODELO hint ('preguntar profesión si no se conoce, luego compartir PDF'). The taller_en_vivo flow (skip fork, one confirmation, pitch + permission + WA group link) is documented in system.md Fase 1 exception (d) and keywords.json hints for TALLER/INTERVENCIONES INTEGRALES/PROCESOS DE ALTO IMPACTO. Examples (Ej10, Ej2b) demonstrate both rules operating correctly in their respective preconditions. No shared precondition produces conflicting outputs.

All examples use consistent Colombian Spanish (tú/tienes/eres, no voseo), warm and collegial tone, short messages (max 2-3 lines), one idea per message, and natural phrasing ('Cuentame', 'Te entiendo', 'Que gusto'). No jarring tonal shifts between examples. Voice matches the Voice section's stated style.

All risky actions in system.md's flow include inline guards. Booking link is guarded by three conditions: (a) know the lead (2-3 exchanges), (b) pitch + alignment + 'esto resuena?', (c) explicit 'deseas agendarla?' + clear yes. PDF delivery is guarded by profession known or keyword context. Derivation to Adri is guarded by family/parent identification. NO_RESPONSE is guarded by explicit triggers (disinterest, emojis alone, minor, etc.). Examples (Ej1, Ej2, Ej6) demonstrate guards firing correctly.

Examples use current flow language and CTAs. Booking CTA matches system.md ('deseas agendarla?', 'aqui puedes agendar'). Taller en vivo CTA matches current promotion (21-may, grupo de WA, 'oferta de lanzamiento'). PDF delivery matches current resource names (recurso_profesion). No stale phrases from prior flow versions detected.

Each behavioral instruction has a single canonical home. The fork familia/profesional is stated once in Fase 1 with exceptions listed. The three booking conditions are stated once in Fase 4. NO_RESPONSE triggers are stated once in the NO_RESPONSE section. Keywords.json hints cross-reference system.md logic without restating it verbatim. No instruction appears in two places with similar prose.

Behavioral guidance (warm tone, short messages, one idea per message, cálida y cercana) is expressed via Voice section + examples, not as rules. Hard rules remain for factual constraints (booking URL, Adri's WhatsApp, no prices in chat, no invention). The four Principles are safety boundaries (no invention, no promises, minors, one close attempt) — appropriate for principles. Tone and conversational style are taught through examples.