# R&Duck v1.0.3 — Progressive Loading Index
# Scan this first (~50 tokens/section). Fetch only what the task needs. AD-04: ≤150-200 instructions.

## Core (always loaded)
/core/boot.md          — activation, session profile, autonomy levels, vocabulary
/core/rules.md         — Golden Rule Index (invariants only, single policy source)

## Core (load as needed)
/core/runtime.md       — freshness, confidence, persistence, re-anchoring, state blocks
/core/routing.md       — domain composition, model routing, saturation, offload
/core/continuity.md    — handoff schema, PSCM corrections, summary packets
/core/review.md        — DA / SPAR / BENCH review ladder

## Capabilities (load for task)
/capabilities/write.md | research.md | audit.md | strategize.md | code.md | rred.md

## Domains (load ONE at a time)
/domains/research | claims-disputes | crisis-response | public-communication
/domains/legal-strategy | technical-analysis | business-strategy | creative-production

## Libraries (loaded by capabilities/domains)
/libraries/personas.md | audiences.md | registers.md

## Workers & Specs
/workers/worker-templates.md    — 7 types + Summary Packet schema
/specs/governance-gate.md       — G25 gate + component registry + DUCK_BUILD

## Research
/research/evolution-ledger.md | decisions.md | feedback.md

## Top-level
/AGENTS.md | CLAIM_BOUNDARY.md | README.md | WIKI.md | CHANGELOG.md

version: 1.0.3 | updated: 2026-06 | license: MIT | repo: github.com/MShneur/R-Duck

## One-Shot Full Load
/llms-full.txt  — ALL core + capability protocols concatenated (~67KB, ~1700 lines)
                   For AIs that can only fetch once per turn, this loads the full system.

## Live Site
https://rd.mstep.org  — all files served as raw markdown (no HTML conversion)
https://mshneur.github.io/R-Duck — mirror (same content) when the custom domain is unavailable