v1.1.0

[1.1.0] — 2026-04-18

Context Efficiency Release. SHINE 1.0 auto-enabled all 16 plugins on install, pushing baseline context to ~95k — combined with user-configured MCPs and a large CLAUDE.md, many sessions crossed Claude Code's effective token budget and became unusable. 1.1 introduces context profiles, slims the always-loaded instruction file, and makes plugin activation opt-in and switchable.

Added

• Context profile system (shine/profiles/*.json) — 6 curated bundles declaring enabledPlugins + disabledMcpjsonServers + estimatedContextTokens:
  • minimal (~15k) — no plugins. New default.
  • writing (~20k) — context7.
  • outbound (~35k) — context7 + prospecting MCPs kept (Apollo, Gmail, CommonRoom).
  • seo (~40k) — context7 + Ahrefs kept.
  • dev (~70k) — full engineering stack (serena, LSPs, playwright, supabase, code-simplifier, superpowers).
  • full (~95k) — all 16 plugins (parity with previous default).
• shine-profile.cjs — Node CLI that atomically rewrites ~/.claude/settings.json's enabledPlugins and disabledMcpjsonServers from a selected profile. Commands: list, show <name>, activate <name>, current. Atomic via fs.renameSync from .tmp.
• shine bash shim (shine/bin/shine) — unified CLI dispatching to shine-profile.cjs (activate|list|show|current|profile) and shine-tools.cjs (doctor|onboard|help|index-skills|--version). Add ~/.claude/shine/bin to PATH to call shine directly.
• install.sh --profile=<name> — CLI flag and interactive picker (6 options, default minimal). Installer now parses the selected profile JSON via node to determine which plugin@marketplace entries to install — instead of the fixed 16-plugin loop.
• shine/references/integrations-map.md — full task-to-integration routing (was CLAUDE.md §§1-14). Loaded on demand.
• shine/references/mcp-capability-map.md — BYO MCP inventory (was CLAUDE.md §15).
• shine/references/agency-playbook.md — tone switching, GDPR guard, proposal assembly, lead enrichment (was CLAUDE.md §§17-20).
• shine/references/tool-tiers.md — free-first fallback rule (was CLAUDE.md §21).
• shine/references/decision-rules.md — full 29-rule matrix (was CLAUDE.md Decision Rules).

Changed

• settings.template.json — the key fix. enabledPlugins now defaults to {} (was 16 entries = true), alwaysThinkingEnabled flipped to false, disabledMcpjsonServers added with explanatory comment. Baseline context drops from ~95k to ~15k.
• CLAUDE.md — split from 31,798 bytes to ~6,300 bytes (80% reduction). Slim core retains identity, auto-use principle, top-level decision rules, and the full §16 Factual / RAG Discipline table (unchanged, MANDATORY). Detailed routing, MCP maps, agency playbooks, and the 29-rule matrix moved to shine/references/*.md for on-demand loading.
• install.sh — profile-aware install flow. Copies shine-profile.cjs, the shine shim, and shine/profiles/*.json; runs shine-profile activate at end of install to sync settings.json with the chosen profile. Help text and next-steps output document the shine activate command.
• shine/VERSION — bumped 1.0.3 → 1.1.0.
• README.md — new "Context profiles" subsection under Quick install, --profile flag documented in flags table, plugin section now describes profile-gated install.
• QUICKSTART.md — install step renumbered and extended with profile picker, new step 3 "Switch context profile", troubleshooting rows for heavy-context symptoms and missing MCPs.

Fixed

• hooks/shine-check-update.js — corrected hardcoded repo slug from the stale diShine-digital-agency/shine-claude-code-framework to diShine-digital-agency/SHINE-Code-System. Previously every update-check silently 404'd.

Migration from 1.0.x

1. Pull the latest SHINE-Code-System and re-run ./install.sh — your existing ~/.claude/ is backed up atomically before any change.
2. Pick a profile when prompted (suggestion: minimal for marketing work, dev for coding, outbound for prospecting).
3. Switch any time with shine activate <name> and restart Claude Code.
4. No data loss: your memory/, MEMORY.md, and custom skills/agents are preserved.

Why this matters

The SHINE 1.0 default combined with a typical user's claude.ai MCP server set easily crossed 200k tokens of context before the first prompt. On Haiku 4.5 sessions with a 200k window, this meant repeated compaction, missing tool schemas, and degraded answer quality. The 1.1 profile system restores headroom: a minimal SHINE install + 5 user MCPs now sits comfortably under 40k, leaving 160k+ for actual work.