MetodologIA Proposal Engine — v4.0

Mental model

This engine has three layers that always run in order:

1. Discover — read the repo/catalog silently before talking
2. Design — apply the right mode (Standard / Innovation / Catalog Edit)
3. Protect — verify legal + brand before generating any file

Never skip layer 3. Never generate files before layer 3 completes.

Layer 1: Discovery (silent — never narrate this to the user)

Run on every activation before responding:

# Load transversal docs (stop at first match per pattern)
for f in 00-inventario-maestro.md 00-glosario-catalogo.md 00-guia-editorial.md \
          00-dod-maestro.md 00-guia-derivacion-versiones.md \
          00-resolucion-por-confirmar.md 00-mapa-interdependencias.md; do
  [ -f "$f" ] && cat "$f"
done

# Load canonical services (head -80 each to get pricing + scope quickly)
find . -name "canonico.md" -not -path "*/.git/*" | while read f; do
  echo "=== $f ===" && head -80 "$f"
done

# Load skill catalog (machine-readable, preferred over .md scan)
[ -f ".claude/skills/metodologia-proposal-engine/catalog/services.yaml" ] && \
  cat .claude/skills/metodologia-proposal-engine/catalog/services.yaml

[ -f ".claude/skills/metodologia-proposal-engine/catalog/conditions.yaml" ] && \
  cat .claude/skills/metodologia-proposal-engine/catalog/conditions.yaml

# Detect unresolved blockers
grep -r "POR CONFIRMAR" --include="*.md" -l 2>/dev/null | head -10

Build a working map:

• Available services + confirmed prices
• Active [POR CONFIRMAR] items (do NOT assert these as confirmed)
• Confirmed credit chains vs. pending ones
• Audience segments available per service

Decision [2026-03-29, JM]: catalog/ YAML files are the preferred SSOT over .md scanning because they are machine-readable and faster. If both exist and conflict, the canonical .md wins. If no repo files exist, fall back to catalog/services.yaml and references/schemas.md.

Layer 2: Design

Mode selection

Incoming request
  → Is the service clearly in the catalog?
      YES → STANDARD MODE
      NO / PARTIALLY →
          Does it fit a known service type (Workshop/Bootcamp/CW/Program)?
              YES → INNOVATION MODE (variant)
              NO  → INNOVATION MODE (new) + [POR CONFIRMAR] on all pricing
  → Does the user want to add/modify a service or condition?
      YES → CATALOG EDIT MODE (read hooks/on-catalog-update.md)

STANDARD MODE

1. Select the canonical service that best matches the need.

2. Read that service's canonico.md fully (not just the head).

3. Determine audience version using catalog/segments.yaml (see also references/brand-system.md for branding mode):

   • B2B executive → ejecutiva-b2b.md pattern: formal 3rd-person, business case, ≤25 words/sentence
   • B2C executive → ejecutiva-b2c.md: direct 2nd-person, personal ROI, ≤25 words
   • B2B commercial → comercial-cliente-b2b.md: narrative hook, benefits, CTA
   • B2C commercial → comercial-cliente-b2c.md: aspirational + concrete, 2nd person
   • Procurement → comercial-compras.md: specs, RACI, SLA, full conditions, no marketing
   • Novice user → usuario-inexperto.md: ≤15 words/sentence, zero jargon, see substitution table

4. Structure content with Minto Complete (see references/voice-content.md):

   • CONCLUSION — decisory recommendation, 1–2 sentences
   • SUPPORT 1 [P1 — (R)Evolution] — gap analysis + method
   • SUPPORT 2 [P2 — Intention over intensity] — design, focus, trade-offs
   • SUPPORT 3 [P3 — Technology as ally] — reusable assets, systems
   • EVIDENCE per support (real data | suggested indicator | measurable signal | required data)
   • CTA — verb + object + context (when/where/who)

5. Resolve branding: call scripts/brand-resolver.js with the brand_mode input (own | cobrand | whitelabel). See references/brand-system.md for full spec. Assumption [low risk]: if no brand_mode is provided, default to own.

INNOVATION MODE

Read references/service-innovation.md before proceeding.

Key rule: new services are composed from canonical building blocks, not invented from scratch. Every module, deliverable, and duration range must trace back to an existing canonical or be explicitly marked [POR CONFIRMAR].

Allowed service types: Workshop (2–4h) | Bootcamp (12–24h) | Consultive Workshop (6–12 × 2h sessions) | Program (8–20 weeks). No new types without JM sign-off.

Required disclaimer in all Innovation Mode outputs:

This is a custom service design proposal. Prices marked [POR CONFIRMAR] are indicative and confirmed before SOW signing. This document does not constitute a delivery commitment or contract.

CATALOG EDIT MODE

Read hooks/on-catalog-update.md. It contains the cascade protocol: which derived files to update, in what order, within what SLA (48h).

Layer 3: Legal + Brand Verification (MANDATORY — do not skip)

Run scripts/verify-legal.js on the full content, or execute the checklist manually if the script is unavailable.

Read references/legal-guardrails.md for the full spec. Summary of hard blockers:

For warnings W1–W7 (co-branding, presential, >20 participants, LMS limits, etc.) see references/legal-guardrails.md#warnings.

Produce a verification report before generating any file:

VERIFICATION REPORT
===================
Status: ✅ APPROVED | ⚠️ APPROVED WITH WARNINGS | 🔴 BLOCKED
Blockers fixed: [list or "none"]
Active warnings: [list or "none"]
Brand mode: [own | cobrand | whitelabel]
Partner brand applied: [name or "n/a"]
Language detected: [en | es | both]
Canonical reference: [slug + version + date]
Mode: [STANDARD | INNOVATION | CATALOG EDIT]

Layer 4: Generation

Only after verification passes. Run scripts/generate-all.js or produce each format manually.

Default output (always, unless user explicitly opts out):

File naming: propuesta_[company-slug]_[YYYY-MM]_[LANG].[ext] Output directory: outputs/ (repo) or /mnt/user-data/outputs/ (Claude Code)

White-label branding rule: when brand_mode: whitelabel, ALL MetodologIA visual identity (logo, colors, name) is replaced with partner brand. Methodology content, structure, and quality standards are unchanged. MetodologIA is completely invisible in the output.

Co-brand rule: both logos appear. MetodologIA design system is the base; partner's primary color is used as an accent override for --gold only. Never override navy or typography.

Layer 5: Delivery

Present files with present_files. Then tell the user:

1. What problem this proposal solves and which service is proposed
2. Any active warnings from verification (if any)
3. All output file paths
4. Any [POR CONFIRMAR] items that must be resolved before sending
5. Offer one iteration: "Anything specific you'd like to adjust?"

Reference map

Load these files only when the situation calls for them:

Hard rules (non-negotiable, every run)

1. Discovery before responding — always.
2. Verification before file generation — always.
3. Never assert a [POR CONFIRMAR] item as confirmed.
4. Never promise a result % without "indicative target" wrapper.
5. "Transformation" → always "(R)Evolution".
6. Prototypes ≠ production-ready automations — always clarify.
7. White-label = MetodologIA completely invisible. Not "subtle branding". Invisible.
8. Co-brand = partner accent only on one token. Never override the design system.
9. IAC is a delivery channel. No IAC-specific code paths. Brand mode handles it.
10. All output is bilingual ES+EN unless user explicitly requests one language only.