specweave 1.0.550 → 1.0.552

package/plugins/specweave/hooks/user-prompt-submit.sh

@@ -1,2550 +0,0 @@
-#!/bin/bash
-
-# SpecWeave UserPromptSubmit Hook (v1.0.533 - Native Plugin Installation)
-# Fires BEFORE user's command executes (prompt-based hook)
-# Purpose: Auto-load plugins, discipline validation, context injection, instant command execution
-#
-# FEATURES:
-# - v1.0.201: LSP CLI FALLBACK INSTRUCTIONS - When LSP requested, instruct Claude to use
-#   `specweave lsp` commands instead of Grep. These use TsServerClient for REAL semantic
-#   analysis. Key fix: "find references" now gets semantic refs, not text matches!
-# - v1.0.198: UNIFIED LLM LSP DETECTION - LLM decides if LSP is needed (replaces grep patterns)
-#   * Single detect-intent call now returns: plugins + increment + skills + LSP recommendation
-#   * LLM-based detection understands context ("find references to X" vs general coding)
-#   * Returns: lsp.needed, lsp.operation (references/definition/hover/symbols), lsp.language
-#   * Grep-based detection kept as fallback if LLM detection not available
-#   * Key insight: LLM can distinguish "find references" request from "build feature" request
-# - v1.0.192: LSP AUTO-INSTALL ON PROJECT DETECTION - Critical fix for LSP plugin installation
-#   * Previously: LSP plugins only installed when user explicitly asked for "find references"
-#   * Now: LSP plugins auto-installed when working on TS/Py/Rust projects
-#   * Project detection: tsconfig.json, package.json, requirements.txt, Cargo.toml, etc.
-#   * Prompt detection: mentions of typescript, react, python, django, rust, etc.
-#   * Supported: vtsls (TS), pyright (Python), rust-analyzer (Rust)
-#   * Key insight: Users shouldn't need to know about LSP to benefit from it
-# - v1.0.177: SKILL CHAINING REMINDER - Add explicit guidance in SKILL FIRST message
-#   * "SKILL FIRST" does NOT mean "only one skill"
-#   * Shows domain skills to use after sw:increment
-#   * Points to CLAUDE.md "MANDATORY: Skill Chaining" section
-# - v1.0.175: CRITICAL FIX - Use installed_plugins.json as SOURCE OF TRUTH
-#   * Reads ~/.claude/plugins/installed_plugins.json directly (eliminates false restart warnings)
-#   * `claude plugin list` can have timing/buffering issues → unreliable for detection
-#   * Primary: check_plugin_installed_from_json() using jq (fast, accurate)
-#   * Fallback: `claude plugin list` only if jq unavailable
-#   * Post-install verification: re-checks registry after install to confirm success
-#   * Increased timeouts: 5s → 10s for CLI operations (reduces timing issues)
-#   * Guard against false positives: if install says "success" but not in registry → treat as already installed
-# - v1.0.169: DIRECT SKILL INVOCATION - Call sw:increment skill directly
-#   * Originally skipped wrapper indirection; increment-planner merged into increment in v1.0.261
-#   * Passes FULL user prompt as args (not just extracted name)
-#   * Uses <system><rules> tags (Claude-trained) instead of custom <mandatory_instruction>
-#   * More concise, imperative instruction text
-# - v1.0.167: FIX PLUGIN RESTART WARNING - Use `claude plugin list` BEFORE install (DEPRECATED - had timing issues)
-#   * Claude CLI always outputs "Successfully installed" even when already installed
-#   * Called `claude plugin list` once to get current plugins, then checked against that
-#   * Skipped install for already-installed plugins (faster + no false restart warnings)
-#   * ISSUE: `claude plugin list` output unreliable → false positives → fixed in v1.0.175
-# - v1.0.147: SYNC PLUGIN INSTALL - Plugins available for CURRENT prompt!
-#   * Replaced 20s async LLM detection with ~200ms sync `claude plugin install`
-#   * Claude Code hot-reload picks up plugins immediately
-#   * Keywords: react, vue, kubernetes, docker, terraform, github, jira, etc.
-#   * Controlled by SPECWEAVE_DISABLE_AUTO_LOAD env var
-# - v1.0.127: (DEPRECATED) Background async plugin detection - too slow, plugins only for next prompt
-# - v1.0.166: CRITICAL FIX - Use hookSpecificOutput.additionalContext (NOT systemMessage!)
-#   * Claude Code ignores systemMessage field in UserPromptSubmit hooks
-#   * Use {"hookSpecificOutput":{"hookEventName":"UserPromptSubmit","additionalContext":"..."}}
-#   * See: https://docs.claude.com/en/docs/claude-code/hooks
-# - v1.0.106: Use approve decision for info commands (not block)
-#   * "block" erases command from context and stops execution
-#   * Info commands (/sw:progress, /sw:status, /sw:jobs, etc.) now use "approve"
-#   * Validation errors (task limit) still use "block" correctly
-# - v1.0.105: Fix ARGS extraction for prompts with IDE metadata prefix (<ide_opened_file>)
-# - v0.34.0: Unified "block" decision for both CLI and VSCode (WRONG - fixed in v1.0.106)
-# - v0.33.1: Unified instant execution - scripts run in hook for both CLI and VSCode
-# - v0.33.0: Script delegation pattern (now deprecated in favor of block decision)
-# - v0.26.13: jq for JSON parsing (10x faster than node -e)
-# - Single active increment detection (cached, not 4x!)
-# - Deferred heavy checks (SpecSyncManager only when needed)
-# - Ultra-fast early exits
-#
-# Performance: Status commands <100ms (was 3+ min), other prompts <10ms
-#
-# ARCHITECTURE:
-# - Both CLI and VSCode: Uses "block" decision with "reason" field to display output and stop execution
-# - Scripts execute in hook - NO LLM involvement for instant commands
-# - Use hookSpecificOutput.additionalContext for UserPromptSubmit (systemMessage is ignored!)
-# - See output_approve_with_context() helper function
-
-set +e
-
-# ==============================================================================
-# ULTRA-FAST EARLY EXIT (before ANY processing)
-# ==============================================================================
-INPUT=$(cat 2>/dev/null || echo '{}')
-
-# Use jq if available (10x faster than node), fallback to portable sed
-if command -v jq >/dev/null 2>&1; then
-  PROMPT=$(echo "$INPUT" | jq -r '.prompt // ""' 2>/dev/null || echo "")
-else
-  # Fallback: extract prompt with sed (portable - works on macOS BSD and Linux GNU)
-  # Note: grep -oP (PCRE) is NOT available on macOS default BSD grep
-  PROMPT=$(echo "$INPUT" | sed -n 's/.*"prompt"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' 2>/dev/null || echo "")
-fi
-
-# ==============================================================================
-# EARLY EXIT FOR BUILT-IN SLASH COMMANDS (v1.0.280, fixed v1.0.305)
-# ==============================================================================
-# Claude Code has built-in slash commands (/context, /help, /clear, /compact,
-# /memory, /permissions, /cost, /doctor, /login, /logout, /config, etc.)
-# SpecWeave MUST NOT intercept, delay, or inject context into these commands.
-# Only /sw: and /sw-*: prefixed commands belong to SpecWeave.
-#
-# Without this guard, built-in commands like /context go through the LLM
-# detect-intent pipeline (5-15s delay) and may get incorrect additionalContext
-# injected, causing them to fail or behave unexpectedly.
-#
-# v1.0.305: Strip IDE metadata tags before checking. In VSCode, the prompt may
-# have <ide_opened_file>...</ide_opened_file> or <ide_selection>...</ide_selection>
-# prefixed on the same line as the command, causing the ^-anchored regex to fail.
-# Uses sed to strip everything up to the last closing </ide_*> tag (handles content
-# with < chars like code selections), then trims leading whitespace.
-CLEAN_PROMPT=$(echo "$PROMPT" | sed 's/.*<\/ide_[a-z_]*>//; s/^[[:space:]]*//')
-if echo "$CLEAN_PROMPT" | grep -qE "^[[:space:]]*/[a-zA-Z][a-zA-Z0-9_-]*($|[[:space:]])" && \
-   ! echo "$CLEAN_PROMPT" | grep -qiE "^[[:space:]]*/sw[-:]"; then
-  echo '{"decision":"approve"}'
-  exit 0
-fi
-
-# ==============================================================================
-# PROJECT-SCOPE INITIALIZATION GUARD (v1.0.235)
-# ==============================================================================
-# Prevents SpecWeave skills from running in non-initialized projects.
-# Provides user-friendly prompt to initialize or disable plugins.
-#
-# Control:
-# - SPECWEAVE_DISABLE_GUARD=1 environment variable
-# - guard.enabled: false in .specweave/config.json
-#
-# Why: Skills are globally visible due to Claude Code's plugin system,
-# but they require project-specific initialization to work correctly.
-
-# Helper function: Find .specweave/config.json by walking up directory tree
-find_specweave_config() {
-  local dir="$PWD"
-  while [[ "$dir" != "/" ]]; do
-    if [[ -f "$dir/.specweave/config.json" ]]; then
-      # Validate it's valid JSON
-      if command -v jq >/dev/null 2>&1; then
-        if jq empty "$dir/.specweave/config.json" 2>/dev/null; then
-          echo "$dir/.specweave/config.json"
-          return 0
-        fi
-      else
-        # No jq available - assume valid if file exists
-        echo "$dir/.specweave/config.json"
-        return 0
-      fi
-    fi
-    dir=$(dirname "$dir")
-  done
-  return 1
-}
-
-# Check if this is a SpecWeave skill invocation
-# Matches: /sw:*, /sw-github:*, /backend:*, /testing:*, etc.
-DOMAIN_PLUGIN_PATTERN="^[[:space:]]*/([Bb]ackend|[Tt]esting|[Mm]obile|[Ii]nfra|[Kk]8s|[Mm]l|[Pp]ayments|[Kk]afka|[Cc]onfluent|[Cc]ost|[Dd]ocs|[Ss]ecurity|[Ss]kills|[Bb]lockchain):[a-zA-Z-]+"
-if [[ "$PROMPT" =~ ^[[:space:]]*/[Ss][Ww](-[a-zA-Z0-9-]+)?:[a-zA-Z-]+ ]] || [[ "$PROMPT" =~ $DOMAIN_PLUGIN_PATTERN ]]; then
-  # Check if guard is disabled via environment variable
-  if [[ "${SPECWEAVE_DISABLE_GUARD:-0}" != "1" ]]; then
-    # Check if project is initialized (walk up tree to find .specweave/config.json)
-    FOUND_CONFIG=$(find_specweave_config)
-    if [[ -z "$FOUND_CONFIG" ]]; then
-      # Check if guard is disabled in config (would fail since no config exists yet)
-      # Extract skill name for error message
-      SKILL_NAME=$(echo "$PROMPT" | grep -oiE '^[[:space:]]*/[a-z0-9-]+:[a-zA-Z-]+' | tr '[:upper:]' '[:lower:]' | sed 's/^[[:space:]]*//')
-
-      # Generate helpful error message
-      cat <<EOF
-{
-  "decision": "block",
-  "reason": "⚠️ **SpecWeave Not Initialized**
-
-You invoked \`${SKILL_NAME}\`, but this project hasn't been initialized with SpecWeave yet.
-
-**Options:**
-
-1. **Initialize SpecWeave in this project:**
-   \`\`\`bash
-   specweave init
-   # or
-   npx specweave init
-   \`\`\`
-
-2. **Use SpecWeave in a different directory:**
-   Navigate to a SpecWeave project first:
-   \`\`\`bash
-   cd /path/to/your/specweave-project
-   \`\`\`
-
-3. **Disable SpecWeave plugins globally:**
-   If you don't need SpecWeave in most projects, disable it in:
-   \`~/.claude/settings.json\`
-   \`\`\`json
-   {
-     \"enabledPlugins\": {
-       \"sw@specweave\": false,
-       \"backend@vskill\": false
-     }
-   }
-   \`\`\`
-
-4. **Disable this guard (not recommended):**
-   \`\`\`bash
-   export SPECWEAVE_DISABLE_GUARD=1
-   \`\`\`
-
-**Note:** SpecWeave skills are globally installed but require project-specific initialization to work correctly."
-}
-EOF
-      exit 0
-    fi
-
-    # Check if guard is disabled in config (using found config path)
-    if command -v jq >/dev/null 2>&1; then
-      GUARD_ENABLED=$(jq -r '.guard.enabled // true' "$FOUND_CONFIG" 2>/dev/null)
-      if [[ "$GUARD_ENABLED" == "false" ]]; then
-        # Guard disabled in config - allow through
-        :
-      fi
-    fi
-  fi
-fi
-
-# ==============================================================================
-# PROJECT ROOT DETECTION (walk up to find .specweave/config.json)
-# ==============================================================================
-# MUST run BEFORE any code that uses SW_PROJECT_ROOT (scope guard, logging, etc.)
-# Checks for config.json to distinguish real projects from stale folders.
-SW_PROJECT_ROOT=""
-_swdir="$PWD"
-while [[ "$_swdir" != "/" ]]; do
-  if [[ -f "$_swdir/.specweave/config.json" ]]; then
-    SW_PROJECT_ROOT="$_swdir"
-    break
-  fi
-  _swdir=$(dirname "$_swdir")
-done
-
-# ==============================================================================
-# USER-LEVEL PLUGIN SCOPE GUARD (v1.0.249)
-# ==============================================================================
-# Prevents SpecWeave domain plugins (sw-*) and LSP plugins (*-lsp) from
-# polluting ~/.claude/settings.json (user scope). These should ALWAYS be
-# project-scoped (.claude/settings.json) to avoid leaking across projects.
-#
-# Sources of user-level pollution:
-# - Claude Code's own plugin discovery (installs at user scope by default)
-# - Older SpecWeave versions (pre-v1.0.210) that didn't enforce project scope
-# - Manual `claude plugin install` without --scope flag
-#
-# What this guard does:
-# 1. Reads installed plugins via `claude plugin list` (fast, <200ms)
-# 2. Identifies sw-*@specweave or *-lsp@* at user scope
-# 3. Uninstalls from user scope and reinstalls at project scope
-# 4. Uses a daily marker file to avoid running on every prompt
-#
-# Rate limit: runs at most once per day per project (marker in .specweave/state/)
-# GUARD: Only run if inside a valid SpecWeave project (prevents stale folder creation)
-SCOPE_GUARD_RUN=false
-SCOPE_GUARD_MARKER=""
-
-if [[ -n "$SW_PROJECT_ROOT" ]]; then
-  SCOPE_GUARD_MARKER="$SW_PROJECT_ROOT/.specweave/state/scope-guard.marker"
-
-  if [[ -f "$SCOPE_GUARD_MARKER" ]]; then
-    # Check if marker is from today (skip if already ran today)
-    MARKER_DATE=$(cat "$SCOPE_GUARD_MARKER" 2>/dev/null)
-    TODAY=$(date +%Y-%m-%d)
-    [[ "$MARKER_DATE" != "$TODAY" ]] && SCOPE_GUARD_RUN=true
-  else
-    SCOPE_GUARD_RUN=true
-  fi
-fi
-
-if [[ "$SCOPE_GUARD_RUN" == "true" ]] && command -v jq >/dev/null 2>&1 && command -v claude >/dev/null 2>&1; then
-  USER_SETTINGS="$HOME/.claude/settings.json"
-
-  if [[ -f "$USER_SETTINGS" ]]; then
-    # Find SpecWeave domain plugins and LSP plugins at user level
-    # Exempt: sw@specweave (core plugin, intentionally user-scoped)
-    POLLUTED_PLUGINS=$(jq -r '
-      .enabledPlugins // {} | to_entries[]
-      | select(
-          (.key | test("^sw-.*@specweave$")) or
-          (.key | test("-lsp@"))
-        )
-      | .key
-    ' "$USER_SETTINGS" 2>/dev/null)
-
-    if [[ -n "$POLLUTED_PLUGINS" ]]; then
-      MIGRATED=""
-      for plugin_key in $POLLUTED_PLUGINS; do
-        # Uninstall from user scope
-        if timeout 5 claude plugin uninstall "$plugin_key" >/dev/null 2>&1; then
-          if [[ "$plugin_key" == sw-*@specweave ]]; then
-            # Reinstall at project scope via native Claude plugin system (v1.0.533)
-            if timeout 10 claude plugin install "$plugin_key" --scope project >/dev/null 2>&1; then
-              [[ -n "$MIGRATED" ]] && MIGRATED="$MIGRATED, "
-              MIGRATED="${MIGRATED}${plugin_key}"
-            fi
-          else
-            # LSP and other plugins: reinstall via claude CLI at project scope
-            if timeout 10 claude plugin install "$plugin_key" --scope project >/dev/null 2>&1; then
-              [[ -n "$MIGRATED" ]] && MIGRATED="$MIGRATED, "
-              MIGRATED="${MIGRATED}${plugin_key}"
-            fi
-          fi
-        fi
-      done
-
-      if [[ -n "$MIGRATED" ]]; then
-        echo "[$(date -Iseconds)] scope-guard | migrated user→project: $MIGRATED" >> "$SW_PROJECT_ROOT/.specweave/state/hook.log" 2>/dev/null || true
-      fi
-
-      # CRITICAL FIX: Restore sw@specweave enabled state after uninstall operations
-      # The `claude plugin uninstall` commands above may corrupt ~/.claude/settings.json
-      # and disable sw@specweave as collateral damage. Re-enable it explicitly.
-      if [[ -f "$USER_SETTINGS" ]]; then
-        SW_ENABLED=$(jq -r '.enabledPlugins."sw@specweave" // "not_set"' "$USER_SETTINGS" 2>/dev/null)
-        if [[ "$SW_ENABLED" != "true" ]]; then
-          # Re-enable core plugin (preserves all other settings)
-          jq '.enabledPlugins."sw@specweave" = true' "$USER_SETTINGS" > "${USER_SETTINGS}.tmp" 2>/dev/null && \
-            mv "${USER_SETTINGS}.tmp" "$USER_SETTINGS" 2>/dev/null || true
-          echo "[$(date -Iseconds)] scope-guard | restored sw@specweave enabled state" >> "$SW_PROJECT_ROOT/.specweave/state/hook.log" 2>/dev/null || true
-        fi
-      fi
-    fi
-  fi
-
-  # v1.0.583: Clean up stale plugins from non-specweave marketplaces (e.g., frontend@vskill).
-  # For each non-exempt marketplace in enabledPlugins, check if it has a valid manifest.
-  # If not, remove all its plugins from settings. Runs daily alongside scope guard.
-  if [[ -f "$USER_SETTINGS" ]]; then
-    STALE_CLEANED=""
-    # Get all unique marketplace names from enabledPlugins (exclude specweave and well-known externals)
-    STALE_MARKETPLACES=$(jq -r '
-      .enabledPlugins // {} | keys[]
-      | split("@")[1] // empty
-      | select(. != "specweave" and . != "claude-plugins-official" and . != "claude-code-lsps")
-    ' "$USER_SETTINGS" 2>/dev/null | sort -u)
-
-    for mkt in $STALE_MARKETPLACES; do
-      MKT_MANIFEST="$HOME/.claude/plugins/marketplaces/$mkt/.claude-plugin/marketplace.json"
-      if [[ ! -f "$MKT_MANIFEST" ]]; then
-        # No manifest — marketplace is dead. Remove ALL its plugins from settings.
-        STALE_KEYS=$(jq -r --arg mkt "$mkt" '
-          .enabledPlugins // {} | keys[] | select(endswith("@" + $mkt))
-        ' "$USER_SETTINGS" 2>/dev/null)
-        for stale_key in $STALE_KEYS; do
-          jq --arg k "$stale_key" 'del(.enabledPlugins[$k])' "$USER_SETTINGS" > "${USER_SETTINGS}.tmp" 2>/dev/null && \
-            mv "${USER_SETTINGS}.tmp" "$USER_SETTINGS" 2>/dev/null || true
-          [[ -n "$STALE_CLEANED" ]] && STALE_CLEANED="$STALE_CLEANED, "
-          STALE_CLEANED="${STALE_CLEANED}${stale_key}"
-        done
-      else
-        # Manifest exists — remove plugins not in the manifest
-        VALID_NAMES=$(jq -r '.plugins[].name' "$MKT_MANIFEST" 2>/dev/null)
-        ENABLED_KEYS=$(jq -r --arg mkt "$mkt" '
-          .enabledPlugins // {} | keys[] | select(endswith("@" + $mkt))
-        ' "$USER_SETTINGS" 2>/dev/null)
-        for key in $ENABLED_KEYS; do
-          plugin_name="${key%%@*}"
-          if ! echo "$VALID_NAMES" | grep -qx "$plugin_name"; then
-            jq --arg k "$key" 'del(.enabledPlugins[$k])' "$USER_SETTINGS" > "${USER_SETTINGS}.tmp" 2>/dev/null && \
-              mv "${USER_SETTINGS}.tmp" "$USER_SETTINGS" 2>/dev/null || true
-            [[ -n "$STALE_CLEANED" ]] && STALE_CLEANED="$STALE_CLEANED, "
-            STALE_CLEANED="${STALE_CLEANED}${key}"
-          fi
-        done
-      fi
-    done
-
-    if [[ -n "$STALE_CLEANED" ]]; then
-      echo "[$(date -Iseconds)] scope-guard | cleaned stale plugins: $STALE_CLEANED" >> "$SW_PROJECT_ROOT/.specweave/state/hook.log" 2>/dev/null || true
-    fi
-  fi
-
-  # Write today's marker
-  mkdir -p "$(dirname "$SCOPE_GUARD_MARKER")" 2>/dev/null
-  date +%Y-%m-%d > "$SCOPE_GUARD_MARKER" 2>/dev/null || true
-fi
-
-# ==============================================================================
-# AUTO-LOAD PLUGIN DETECTION + INCREMENT ASSIST (v1.0.141)
-# ==============================================================================
-# Detect plugin needs AND increment creation suggestions using LLM (Claude Haiku)
-# This runs BEFORE the SpecWeave keyword check to catch plugin-specific prompts
-#
-# Plugin Auto-Load Control:
-# - SPECWEAVE_DISABLE_AUTO_LOAD=1 environment variable
-# - pluginAutoLoad.enabled: false in .specweave/config.json
-#
-# Increment Assist Control:
-# - incrementAssist.enabled: false in .specweave/config.json (disables suggestions)
-# - incrementAssist.confidenceThreshold: 0.7 (minimum confidence to show suggestion)
-#
-# When both disabled: NO detection, NO LLM calls, fastest response time (~5-7s saved)
-
-# PROJECT ROOT DETECTION was moved earlier (before scope guard) to prevent stale folder creation
-
-# Check config for pluginAutoLoad.enabled, suggestOnly and incrementAssist.enabled settings
-PLUGIN_AUTOLOAD_ENABLED=true
-PLUGIN_SUGGEST_ONLY=true
-INCREMENT_ASSIST_ENABLED=true
-INCREMENT_CONFIDENCE_THRESHOLD=0.7
-INCREMENT_MANDATORY_CONFIG=true
-DEEP_INTERVIEW_ENABLED=false
-if [[ -n "$SW_PROJECT_ROOT" ]]; then
-  CONFIG_PATH="$SW_PROJECT_ROOT/.specweave/config.json"
-else
-  CONFIG_PATH=".specweave/config.json"
-fi
-if [[ -f "$CONFIG_PATH" ]]; then
-  if command -v jq >/dev/null 2>&1; then
-    AUTOLOAD_VALUE=$(jq -r '.pluginAutoLoad.enabled // true' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$AUTOLOAD_VALUE" == "false" ]] && PLUGIN_AUTOLOAD_ENABLED=false
-
-    # Check suggestOnly mode (v1.0.158, default flipped to true in v1.0.397 — consent-first)
-    SUGGEST_VALUE=$(jq -r '.pluginAutoLoad.suggestOnly // true' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$SUGGEST_VALUE" == "false" ]] && PLUGIN_SUGGEST_ONLY=false
-
-    INCREMENT_VALUE=$(jq -r '.incrementAssist.enabled // true' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$INCREMENT_VALUE" == "false" ]] && INCREMENT_ASSIST_ENABLED=false
-
-    THRESHOLD_VALUE=$(jq -r '.incrementAssist.confidenceThreshold // 0.7' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$THRESHOLD_VALUE" =~ ^[0-9.]+$ ]] && INCREMENT_CONFIDENCE_THRESHOLD="$THRESHOLD_VALUE"
-
-    # Read incrementAssist.mandatory from config (config-based override for blocking)
-    MANDATORY_CONFIG_VALUE=$(jq -r '.incrementAssist.mandatory // true' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$MANDATORY_CONFIG_VALUE" == "false" ]] && INCREMENT_MANDATORY_CONFIG=false
-
-    # Deep Interview Mode detection (v1.0.195)
-    DEEP_INTERVIEW_VALUE=$(jq -r '.planning.deepInterview.enabled // false' "$CONFIG_PATH" 2>/dev/null)
-    [[ "$DEEP_INTERVIEW_VALUE" == "true" ]] && DEEP_INTERVIEW_ENABLED=true
-  else
-    # Fallback: grep for explicit false settings
-    if grep -q '"pluginAutoLoad"' "$CONFIG_PATH" 2>/dev/null && grep -q '"enabled"[[:space:]]*:[[:space:]]*false' "$CONFIG_PATH" 2>/dev/null; then
-      PLUGIN_AUTOLOAD_ENABLED=false
-    fi
-    # Fallback: grep for explicit suggestOnly=false (opt-in to auto-install)
-    if grep -q '"pluginAutoLoad"' "$CONFIG_PATH" 2>/dev/null && grep -A5 '"pluginAutoLoad"' "$CONFIG_PATH" 2>/dev/null | grep -q '"suggestOnly"[[:space:]]*:[[:space:]]*false'; then
-      PLUGIN_SUGGEST_ONLY=false
-    fi
-    if grep -q '"incrementAssist"' "$CONFIG_PATH" 2>/dev/null && grep -A5 '"incrementAssist"' "$CONFIG_PATH" 2>/dev/null | grep -q '"enabled"[[:space:]]*:[[:space:]]*false'; then
-      INCREMENT_ASSIST_ENABLED=false
-    fi
-    # Fallback: grep for deep interview mode
-    if grep -q '"deepInterview"' "$CONFIG_PATH" 2>/dev/null && grep -A5 '"deepInterview"' "$CONFIG_PATH" 2>/dev/null | grep -q '"enabled"[[:space:]]*:[[:space:]]*true'; then
-      DEEP_INTERVIEW_ENABLED=true
-    fi
-  fi
-fi
-
-# ==============================================================================
-# HELPER FUNCTIONS (must be defined before use)
-# ==============================================================================
-
-# Helper: Escape output for JSON (handles newlines, quotes, backslashes)
-escape_json_early() {
-  local input="$1"
-  if command -v jq >/dev/null 2>&1; then
-    printf '%s' "$input" | jq -Rs '.' | sed 's/^"//; s/"$//'
-  else
-    # Fallback: basic escaping without jq
-    printf '%s' "$input" | sed 's/\\/\\\\/g; s/"/\\"/g' | awk '{printf "%s\\n", $0}' | sed 's/\\n$//'
-  fi
-}
-
-# v1.0.254: Prompt safety limits to prevent "Prompt is too long" errors
-# Must match MAX_ADDITIONAL_CONTEXT_LENGTH in src/core/lazy-loading/llm-plugin-detector.ts
-MAX_ADDITIONAL_CONTEXT_LENGTH=3000
-
-# Helper: Output approve response with context (Claude Code hook format v1.0.166)
-# CRITICAL: systemMessage is NOT a valid field for UserPromptSubmit hooks!
-# Use hookSpecificOutput.additionalContext instead.
-# See: https://docs.claude.com/en/docs/claude-code/hooks
-#
-# v1.0.254: Added size guard — truncates additionalContext if it exceeds
-# MAX_ADDITIONAL_CONTEXT_LENGTH to prevent "Prompt is too long" errors.
-output_approve_with_context() {
-  local context="$1"
-  # v1.0.254: Safety truncation to prevent prompt overflow
-  # v1.0.260: Added overflow logging for debugging context budget issues
-  if [[ ${#context} -gt $MAX_ADDITIONAL_CONTEXT_LENGTH ]]; then
-    local overflow_by=$(( ${#context} - MAX_ADDITIONAL_CONTEXT_LENGTH ))
-    echo "[$(date -Iseconds)] CONTEXT OVERFLOW | size=${#context} | max=$MAX_ADDITIONAL_CONTEXT_LENGTH | overflow_by=$overflow_by | truncating" >> "${LAZY_LOAD_LOG:-/dev/null}" 2>/dev/null
-    context="${context:0:$MAX_ADDITIONAL_CONTEXT_LENGTH}... [context truncated for safety]"
-  fi
-  local escaped
-  escaped=$(escape_json_early "$context")
-  printf '{"hookSpecificOutput":{"hookEventName":"UserPromptSubmit","additionalContext":"%s"}}\n' "$escaped"
-}
-
-# v1.0.260: truncate_and_escape_prompt() removed — prompt embedding in SKILL FIRST
-# was eliminated to save ~800 chars of context budget per turn. The skill reads
-# the user's prompt from conversation context (it's already there).
-
-# Helper: Check if plugin is in vskill.lock (fast-path skip) (v1.0.272)
-# vskill.lock is the SOURCE OF TRUTH for vskill-installed plugins.
-# Args: $1=plugin name (e.g., "backend")
-# Returns: 0 if in lockfile, 1 if not
-check_plugin_in_vskill_lock() {
-  local plugin="$1"
-  local lockfile="vskill.lock"
-
-  # Check project-local lockfile first
-  [[ ! -f "$lockfile" ]] && return 1
-
-  # Must have jq for reliable JSON parsing
-  if ! command -v jq >/dev/null 2>&1; then
-    # Fallback: grep for plugin name in lockfile
-    grep -q "\"${plugin}\"" "$lockfile" 2>/dev/null && return 0
-    return 1
-  fi
-
-  # Check if plugin exists in lockfile skills
-  local has_skill
-  has_skill=$(jq -r --arg key "$plugin" '.skills[$key] // null' "$lockfile" 2>/dev/null)
-
-  if [[ "$has_skill" != "null" ]] && [[ -n "$has_skill" ]]; then
-    return 0  # Already installed via vskill
-  else
-    return 1  # Not in lockfile
-  fi
-}
-
-# ==============================================================================
-# check_plugin_installed_from_json REMOVED (v1.0.535)
-# Plugins are now pre-installed at init time. No runtime installation checks needed.
-# ==============================================================================
-
-# ==============================================================================
-# KEYWORD-BASED PLUGIN DETECTION REMOVED (v1.0.159)
-# ==============================================================================
-# Keyword fallback was removed because it was too aggressive.
-# Example: "run tests" would install testing even for simple test runs.
-#
-# Plugin detection now happens ONLY via LLM analysis (specweave detect-intent).
-# The LLM understands user INTENT - it only recommends plugins when user
-# explicitly asks to BUILD/IMPLEMENT something, not for questions or discussions.
-
-# ==============================================================================
-# UNIFIED LLM DETECTION (v1.0.147) - ONE call for BOTH plugins AND increments
-# ==============================================================================
-# Single `specweave detect-intent` call returns:
-# - plugins: which plugins to install (for CURRENT prompt via hot-reload)
-# - increment: whether to suggest creating an increment
-#
-# DEFAULT: SUGGEST INCREMENT for ~95% of implementation work.
-# LLM decides action: "new" (most cases), "small_fix" (trivial edits),
-# "hotfix" (urgent), "reopen" (related to previous), "none" (skip).
-#
-# WHEN NOT TO CREATE INCREMENT (action: "none" ONLY):
-# ┌─────────────────────────────────────────────────────────────────────────────┐
-# │ • Pure questions: "what is X?", "explain Y", "tell me about Z"            │
-# │ • Exploration: "show me", "list", "search for"                            │
-# │ • Commands: "run tests", "build", "deploy", "commit"                      │
-# │ • Already in workflow: prompt starts with /sw:                             │
-# │ • Chat/greetings: "hello", "thanks", general conversation                 │
-# │ • EXPLICIT OPT-OUT: "don't create an increment", "skip workflow"          │
-# │                                                                            │
-# │ NOTE: These are NOT questions — they are WORK requiring increments:        │
-# │   "investigate", "debug", "troubleshoot", "why does X fail",              │
-# │   "optimize", "improve", "secure", "audit", "solve", "resolve",          │
-# │   "analyze", "root cause", "X is broken", "X keeps failing"              │
-# └─────────────────────────────────────────────────────────────────────────────┘
-#
-# STILL SUGGEST INCREMENT (action: "small_fix"):
-# │ • Typo fixes, version bumps, single config value changes                   │
-# │ • These get a non-mandatory suggestion so user can opt in                  │
-
-# Initialize message variable
-AUTOLOAD_PLUGINS_MSG=""
-
-# ==============================================================================
-# EXTERNAL FOLDER DETECTION (v1.0.166) - Quick recommendation
-# ==============================================================================
-# If user wants to create project in EXTERNAL folder, recommend new session
-EXTERNAL_FOLDER_DETECTED=""
-if echo "$PROMPT" | grep -qiE "(in|to|at|create)[[:space:]]+(external[[:space:]]+folder|separate[[:space:]]+folder|new[[:space:]]+folder|~/|/Users/|/home/|/tmp/|outside[[:space:]]+(this|current|the)[[:space:]]+project)"; then
-  # Extract the target path if mentioned
-  TARGET_PATH=$(echo "$PROMPT" | grep -oE "~/[a-zA-Z0-9_/-]+" | head -1)
-  [[ -z "$TARGET_PATH" ]] && TARGET_PATH=$(echo "$PROMPT" | grep -oE "/Users/[a-zA-Z0-9_/-]+" | head -1)
-  [[ -z "$TARGET_PATH" ]] && TARGET_PATH=$(echo "$PROMPT" | grep -oE "/home/[a-zA-Z0-9_/-]+" | head -1)
-
-  EXTERNAL_FOLDER_DETECTED="📁 **EXTERNAL PROJECT DETECTED**
-
-You're requesting work in a folder outside this project.
-${TARGET_PATH:+Target: $TARGET_PATH}
-
-💡 **Recommended**: Start a NEW Claude Code session in that folder:
-   1. Open terminal: \`cd ${TARGET_PATH:-<target-folder>}\`
-   2. Start Claude Code: \`claude\`
-   3. Run \`specweave init\` to set up SpecWeave there
-
-This ensures plugins and context are properly scoped to the new project.
-
----
-
-"
-fi
-
-# ==============================================================================
-# LSP LANGUAGE SERVER CHECK (v1.0.179) - One-time warning for missing servers
-# ==============================================================================
-# Reads results from background lsp-check.sh (spawned by session-start.sh)
-# Shows warning ONCE per session if language servers are missing
-LSP_WARNING_MSG=""
-if [[ -n "$SW_PROJECT_ROOT" ]]; then
-  LSP_STATE_FILE="$SW_PROJECT_ROOT/.specweave/state/lsp-check.json"
-else
-  LSP_STATE_FILE=""
-fi
-if [[ -f "$LSP_STATE_FILE" ]] && command -v jq >/dev/null 2>&1; then
-  LSP_STATUS=$(jq -r '.status // "ok"' "$LSP_STATE_FILE" 2>/dev/null)
-  LSP_WARNED=$(jq -r '.warned // false' "$LSP_STATE_FILE" 2>/dev/null)
-
-  if [[ "$LSP_STATUS" == "missing" ]] && [[ "$LSP_WARNED" != "true" ]]; then
-    # Build warning message from missing servers
-    MISSING_SERVERS=$(jq -r '.missing[] | "- **\(.language)**: `\(.install)`"' "$LSP_STATE_FILE" 2>/dev/null)
-
-    if [[ -n "$MISSING_SERVERS" ]]; then
-      LSP_WARNING_MSG="LSP missing: ${MISSING_SERVERS}. Install for semantic code intelligence. Guide: https://verified-skill.com/docs/guides/lsp-integration
-
-"
-      # Mark as warned so we don't show again this session
-      # Use a temp file to avoid jq in-place issues
-      TMP_FILE=$(mktemp)
-      jq '.warned = true' "$LSP_STATE_FILE" > "$TMP_FILE" 2>/dev/null && mv "$TMP_FILE" "$LSP_STATE_FILE" 2>/dev/null
-    fi
-  fi
-fi
-
-# ==============================================================================
-# LSP PROJECT CONFIG (v1.0.192) - Project-level LSP configuration
-# ==============================================================================
-# Reads LSP settings from .specweave/config.json instead of requiring env vars
-# Config schema:
-#   lsp.enabled: true/false - Enable LSP features for this project
-#   lsp.autoInstallPlugins: true/false - Auto-install marketplace and plugins
-#   lsp.marketplace: "boostvolt/claude-code-lsps" - Which marketplace to use
-#   lsp.plugins.typescript: "vtsls" - TypeScript LSP plugin
-#   lsp.plugins.python: "pyright" - Python LSP plugin
-
-LSP_CONFIG_ENABLED="false"
-LSP_AUTO_INSTALL="false"
-LSP_MARKETPLACE="boostvolt/claude-code-lsps"
-LSP_MARKETPLACE_URL="https://github.com/boostvolt/claude-code-lsps"
-LSP_ENV_WARNED="false"
-
-if [[ -f "$CONFIG_PATH" ]] && command -v jq >/dev/null 2>&1; then
-  LSP_CONFIG_ENABLED=$(jq -r '.lsp.enabled // false' "$CONFIG_PATH" 2>/dev/null)
-  LSP_AUTO_INSTALL=$(jq -r '.lsp.autoInstallPlugins // false' "$CONFIG_PATH" 2>/dev/null)
-  LSP_MARKETPLACE=$(jq -r '.lsp.marketplace // "boostvolt/claude-code-lsps"' "$CONFIG_PATH" 2>/dev/null)
-fi
-
-# Check for LSP request keywords (find references, go to definition, etc.)
-# v1.0.198: This is now a FALLBACK - LLM detection (lsp.needed field) takes precedence
-# The LLM-based detection in detect-intent is more accurate and understands context
-LSP_REQUEST_DETECTED="false"
-if echo "$PROMPT" | grep -qiE "(find|get|show)[[:space:]]+(all[[:space:]]+)?references|go[[:space:]]?to[[:space:]]?definition|goto[[:space:]]?definition|LSP|findReferences|goToDefinition|hover|documentSymbol|workspaceSymbol"; then
-  LSP_REQUEST_DETECTED="true"
-fi
-
-# ==============================================================================
-# LSP PROJECT LANGUAGE DETECTION (v1.0.192) - Auto-detect project languages
-# ==============================================================================
-# Detects project languages from file system to auto-install LSP plugins
-# LAZY LOADING: Only installs when project/prompt actually needs that language
-# Available plugins in boostvolt/claude-code-lsps marketplace:
-#   - vtsls: TypeScript/JavaScript (most common)
-#   - pyright: Python
-#   - rust-analyzer: Rust
-# Key insight: User working on TS project should get LSP without asking for it
-LSP_PROJECT_NEEDS_TS="false"
-LSP_PROJECT_NEEDS_PY="false"
-LSP_PROJECT_NEEDS_RUST="false"
-LSP_PROJECT_NEEDS_CSHARP="false"
-LSP_PROJECT_NEEDS_GO="false"
-LSP_PROJECT_NEEDS_JAVA="false"
-LSP_PROMPT_NEEDS_TS="false"
-LSP_PROMPT_NEEDS_PY="false"
-LSP_PROMPT_NEEDS_RUST="false"
-LSP_PROMPT_NEEDS_CSHARP="false"
-LSP_PROMPT_NEEDS_GO="false"
-LSP_PROMPT_NEEDS_JAVA="false"
-
-# Detect TypeScript/JavaScript project from file system
-# Check for: tsconfig.json, package.json with typescript, *.ts/*.tsx files
-if [[ -f "tsconfig.json" ]] || [[ -f "tsconfig.base.json" ]] || [[ -f "jsconfig.json" ]]; then
-  LSP_PROJECT_NEEDS_TS="true"
-elif [[ -f "package.json" ]]; then
-  # Check if package.json mentions typescript
-  if grep -qE '"typescript"|"@types/|"tsx"|"ts-node"' package.json 2>/dev/null; then
-    LSP_PROJECT_NEEDS_TS="true"
-  fi
-fi
-# Also check for .ts/.tsx files in src/ or root (fast check, max 1 level deep)
-if [[ "$LSP_PROJECT_NEEDS_TS" != "true" ]]; then
-  if ls *.ts *.tsx src/*.ts src/*.tsx 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_TS="true"
-  fi
-fi
-
-# Detect Python project from file system
-# Check for: requirements.txt, pyproject.toml, setup.py, *.py files
-if [[ -f "requirements.txt" ]] || [[ -f "pyproject.toml" ]] || [[ -f "setup.py" ]] || [[ -f "Pipfile" ]]; then
-  LSP_PROJECT_NEEDS_PY="true"
-fi
-if [[ "$LSP_PROJECT_NEEDS_PY" != "true" ]]; then
-  if ls *.py src/*.py 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_PY="true"
-  fi
-fi
-
-# Detect Rust project from file system
-# Check for: Cargo.toml, Cargo.lock, *.rs files
-if [[ -f "Cargo.toml" ]] || [[ -f "Cargo.lock" ]]; then
-  LSP_PROJECT_NEEDS_RUST="true"
-fi
-if [[ "$LSP_PROJECT_NEEDS_RUST" != "true" ]]; then
-  if ls *.rs src/*.rs 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_RUST="true"
-  fi
-fi
-
-# v1.0.235: Detect C#/.NET project from file system
-# Check for: *.csproj, *.sln, Directory.Build.props, *.cs files
-if ls *.csproj *.sln 2>/dev/null | head -1 | grep -q . || [[ -f "Directory.Build.props" ]]; then
-  LSP_PROJECT_NEEDS_CSHARP="true"
-fi
-if [[ "$LSP_PROJECT_NEEDS_CSHARP" != "true" ]]; then
-  # Check one level of subdirectories (common C# project structure: src/MyApp/MyApp.csproj)
-  if ls */*.csproj */*.sln src/*/*.csproj 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_CSHARP="true"
-  elif ls *.cs src/*.cs 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_CSHARP="true"
-  fi
-fi
-
-# v1.0.235: Detect Go project from file system
-if [[ -f "go.mod" ]] || [[ -f "go.sum" ]]; then
-  LSP_PROJECT_NEEDS_GO="true"
-fi
-if [[ "$LSP_PROJECT_NEEDS_GO" != "true" ]]; then
-  if ls *.go cmd/*.go pkg/*.go 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_GO="true"
-  fi
-fi
-
-# v1.0.235: Detect Java project from file system
-if [[ -f "pom.xml" ]] || [[ -f "build.gradle" ]] || [[ -f "build.gradle.kts" ]] || [[ -f "settings.gradle" ]]; then
-  LSP_PROJECT_NEEDS_JAVA="true"
-fi
-if [[ "$LSP_PROJECT_NEEDS_JAVA" != "true" ]]; then
-  if ls *.java src/*.java src/main/java/*.java 2>/dev/null | head -1 | grep -q .; then
-    LSP_PROJECT_NEEDS_JAVA="true"
-  fi
-fi
-
-# Detect from prompt keywords (TypeScript/React/Vue/Angular/Node)
-if echo "$PROMPT" | grep -qiE "\.tsx?|typescript|react|vue|angular|next\.?js|node\.?js|express|nestjs"; then
-  LSP_PROMPT_NEEDS_TS="true"
-fi
-
-# Detect from prompt keywords (Python/Django/Flask/FastAPI)
-if echo "$PROMPT" | grep -qiE "\.py|python|django|flask|fastapi|pytorch|tensorflow|pandas|numpy"; then
-  LSP_PROMPT_NEEDS_PY="true"
-fi
-
-# Detect from prompt keywords (Rust/Cargo)
-if echo "$PROMPT" | grep -qiE "\.rs|rust|cargo|rustc|tokio|actix|axum"; then
-  LSP_PROMPT_NEEDS_RUST="true"
-fi
-
-# v1.0.235: Detect from prompt keywords (C#/.NET)
-if echo "$PROMPT" | grep -qiE "\.cs\b|c#|csharp|dotnet|\.net|asp\.net|blazor|entity.?framework|nuget|\.csproj|\.sln"; then
-  LSP_PROMPT_NEEDS_CSHARP="true"
-fi
-
-# v1.0.235: Detect from prompt keywords (Go)
-if echo "$PROMPT" | grep -qiE "\.go\b|golang|go\.mod|goroutine|gin|echo|fiber"; then
-  LSP_PROMPT_NEEDS_GO="true"
-fi
-
-# v1.0.235: Detect from prompt keywords (Java/Spring/Kotlin)
-if echo "$PROMPT" | grep -qiE "\.java\b|java\b|spring|maven|gradle|kotlin|\.kt\b|jvm"; then
-  LSP_PROMPT_NEEDS_JAVA="true"
-fi
-
-# LSP setup is handled by `specweave init` and `specweave lsp status`
-# No per-prompt warnings needed - avoids state file pollution
-LSP_ENV_SETUP_MSG=""
-
-# ==============================================================================
-# LSP AUTO-INSTALL (v1.0.196) - Install LSP plugins with PROJECT SCOPE
-# ==============================================================================
-# Triggers on ANY of:
-#   - Explicit LSP request (findReferences, goToDefinition, etc.)
-#   - Project language detection (tsconfig.json, package.json, requirements.txt, Cargo.toml)
-#   - Prompt language detection (mentions typescript, react, python, django, etc.)
-# This ensures LSP plugins are installed when working on TS/Py/Rust projects
-# WITHOUT requiring user to explicitly ask for "find references"
-#
-# v1.0.196: LSP plugins now install with PROJECT SCOPE by default
-#   - Reads plugins.scope.lspScope from config (default: "project")
-#   - Project scope keeps LSP plugins specific to the project
-#   - Avoids polluting global plugin list with project-specific language support
-LSP_INSTALL_MSG=""
-LSP_NEEDS_INSTALL="false"
-
-# Read LSP scope from config.json (v1.0.196)
-# Default: "project" - LSP plugins should be project-scoped
-LSP_PLUGIN_SCOPE="project"
-if [[ -f "$CONFIG_PATH" ]] && command -v jq >/dev/null 2>&1; then
-  SCOPE_VALUE=$(jq -r '.plugins.scope.lspScope // "project"' "$CONFIG_PATH" 2>/dev/null)
-  if [[ "$SCOPE_VALUE" == "user" ]] || [[ "$SCOPE_VALUE" == "project" ]] || [[ "$SCOPE_VALUE" == "local" ]]; then
-    LSP_PLUGIN_SCOPE="$SCOPE_VALUE"
-  fi
-fi
-
-# Check if ANY language detection triggered
-if [[ "$LSP_REQUEST_DETECTED" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_CSHARP" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_CSHARP" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_GO" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_GO" == "true" ]] || \
-   [[ "$LSP_PROJECT_NEEDS_JAVA" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_JAVA" == "true" ]]; then
-  LSP_NEEDS_INSTALL="true"
-fi
-
-# ==============================================================================
-# LSP SETUP SUGGESTION (v1.0.203) - Suggest setup instead of auto-installing
-# ==============================================================================
-# When languages are detected but auto-install is disabled, suggest running
-# specweave lsp setup which does multi-repo scanning and asks user approval
-LSP_SETUP_SUGGESTION_MSG=""
-if [[ -n "$SW_PROJECT_ROOT" ]]; then
-  LSP_SETUP_STATE_FILE="$SW_PROJECT_ROOT/.specweave/state/lsp-setup-suggested.flag"
-else
-  LSP_SETUP_STATE_FILE=""
-fi
-
-if [[ "$LSP_NEEDS_INSTALL" == "true" ]] && [[ "$LSP_AUTO_INSTALL" != "true" ]]; then
-  # Check if we've already suggested setup in this session
-  if [[ ! -f "$LSP_SETUP_STATE_FILE" ]] && [[ -n "${ENABLE_LSP_TOOL:-}" ]]; then
-    # Build list of detected languages
-    DETECTED_LANGS=""
-    if [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]]; then
-      DETECTED_LANGS="TypeScript"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]]; then
-      [[ -n "$DETECTED_LANGS" ]] && DETECTED_LANGS="$DETECTED_LANGS, "
-      DETECTED_LANGS="${DETECTED_LANGS}Python"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]]; then
-      [[ -n "$DETECTED_LANGS" ]] && DETECTED_LANGS="$DETECTED_LANGS, "
-      DETECTED_LANGS="${DETECTED_LANGS}Rust"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_CSHARP" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_CSHARP" == "true" ]]; then
-      [[ -n "$DETECTED_LANGS" ]] && DETECTED_LANGS="$DETECTED_LANGS, "
-      DETECTED_LANGS="${DETECTED_LANGS}C#"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_GO" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_GO" == "true" ]]; then
-      [[ -n "$DETECTED_LANGS" ]] && DETECTED_LANGS="$DETECTED_LANGS, "
-      DETECTED_LANGS="${DETECTED_LANGS}Go"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_JAVA" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_JAVA" == "true" ]]; then
-      [[ -n "$DETECTED_LANGS" ]] && DETECTED_LANGS="$DETECTED_LANGS, "
-      DETECTED_LANGS="${DETECTED_LANGS}Java"
-    fi
-
-    # v1.0.235: Check if plugins are missing (expanded language support)
-    MISSING_PLUGINS="false"
-    INSTALLED_PLUGINS_FILE="$HOME/.claude/plugins/installed_plugins.json"
-    _check_plugin() {
-      local plugin_name="$1"
-      if [[ -f "$INSTALLED_PLUGINS_FILE" ]]; then
-        if ! grep -q "\"${plugin_name}@" "$INSTALLED_PLUGINS_FILE" 2>/dev/null; then
-          MISSING_PLUGINS="true"
-        fi
-      else
-        MISSING_PLUGINS="true"
-      fi
-    }
-    if [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]]; then
-      _check_plugin "vtsls"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]]; then
-      _check_plugin "pyright"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]]; then
-      _check_plugin "rust-analyzer"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_CSHARP" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_CSHARP" == "true" ]]; then
-      _check_plugin "csharp-lsp"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_GO" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_GO" == "true" ]]; then
-      _check_plugin "gopls"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_JAVA" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_JAVA" == "true" ]]; then
-      _check_plugin "jdtls"
-    fi
-
-    if [[ "$MISSING_PLUGINS" == "true" ]] && [[ -n "$DETECTED_LANGS" ]]; then
-      LSP_SETUP_SUGGESTION_MSG="💡 **LSP Setup Available**
-
-Detected languages: **${DETECTED_LANGS}**
-
-For enhanced code intelligence (find references, go to definition, hover), run:
-\`\`\`bash
-specweave lsp setup
-\`\`\`
-
-This will scan your project (including nested repos) and let you choose which LSP plugins to install.
-
----
-
-"
-      # Mark as suggested (only in initialized SpecWeave projects)
-      if [[ -n "$LSP_SETUP_STATE_FILE" ]] && [[ -n "$SW_PROJECT_ROOT" ]]; then
-        mkdir -p "$(dirname "$LSP_SETUP_STATE_FILE")" 2>/dev/null
-        touch "$LSP_SETUP_STATE_FILE" 2>/dev/null
-      fi
-    fi
-  fi
-fi
-
-if [[ "$LSP_NEEDS_INSTALL" == "true" ]] && [[ "$LSP_AUTO_INSTALL" == "true" ]]; then
-  # CRITICAL: Skip LSP plugin installation if ENABLE_LSP_TOOL is not set (v1.0.195)
-  # Installing plugins without this env var is useless - they won't work
-  if [[ -z "${ENABLE_LSP_TOOL:-}" ]]; then
-    # Don't install - just show the setup message (already handled by LSP_ENV_SETUP_MSG)
-    LSP_NEEDS_INSTALL="false"
-  fi
-fi
-
-if [[ "$LSP_NEEDS_INSTALL" == "true" ]] && [[ "$LSP_AUTO_INSTALL" == "true" ]]; then
-  # v1.0.397: Respect global suggest-only mode for LSP plugins too (consent-first)
-  if [[ "$PLUGIN_SUGGEST_ONLY" == "true" ]]; then
-    LSP_SUGGEST_CMDS=""
-    if [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]]; then
-      VTSLS_INSTALLED=$(jq -r '."vtsls@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      [[ "$VTSLS_INSTALLED" != "true" ]] && LSP_SUGGEST_CMDS="${LSP_SUGGEST_CMDS}  - **TypeScript**: \`claude plugin install vtsls@claude-code-lsps --scope ${LSP_PLUGIN_SCOPE}\`\n"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]]; then
-      PYRIGHT_INSTALLED=$(jq -r '."pyright@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      [[ "$PYRIGHT_INSTALLED" != "true" ]] && LSP_SUGGEST_CMDS="${LSP_SUGGEST_CMDS}  - **Python**: \`claude plugin install pyright@claude-code-lsps --scope ${LSP_PLUGIN_SCOPE}\`\n"
-    fi
-    if [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]]; then
-      RUST_ANALYZER_INSTALLED=$(jq -r '."rust-analyzer@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      [[ "$RUST_ANALYZER_INSTALLED" != "true" ]] && LSP_SUGGEST_CMDS="${LSP_SUGGEST_CMDS}  - **Rust**: \`claude plugin install rust-analyzer@claude-code-lsps --scope ${LSP_PLUGIN_SCOPE}\`\n"
-    fi
-    if [[ -n "$LSP_SUGGEST_CMDS" ]]; then
-      LSP_INSTALL_MSG="**Suggested LSP plugins**:\n${LSP_SUGGEST_CMDS}After installing, **restart Claude Code** to use LSP features.\n\n---\n\n"
-    fi
-  else
-    # NORMAL MODE (user opted in with suggestOnly: false) - Actually install LSP plugins
-    # Check if marketplace is already installed
-    MARKETPLACE_DIR="$HOME/.claude/plugins/marketplaces/claude-code-lsps"
-    if [[ ! -d "$MARKETPLACE_DIR" ]] && command -v claude >/dev/null 2>&1; then
-      # Install the marketplace
-      if timeout 15 claude plugin marketplace add "$LSP_MARKETPLACE_URL" >/dev/null 2>&1; then
-        LSP_INSTALL_MSG="✅ **LSP marketplace installed**: \`$LSP_MARKETPLACE\`
-"
-      fi
-    fi
-
-    # Auto-install TypeScript LSP plugin (vtsls) when TypeScript project/prompt detected
-    # v1.0.196: Uses --scope $LSP_PLUGIN_SCOPE (default: project)
-    if [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]]; then
-      VTSLS_INSTALLED=$(jq -r '."vtsls@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      if [[ "$VTSLS_INSTALLED" != "true" ]] && command -v claude >/dev/null 2>&1; then
-        if timeout 15 claude plugin install vtsls@claude-code-lsps --scope $LSP_PLUGIN_SCOPE >/dev/null 2>&1; then
-          LSP_INSTALL_MSG="${LSP_INSTALL_MSG}✅ **TypeScript LSP installed**: \`vtsls@claude-code-lsps\` (scope: $LSP_PLUGIN_SCOPE)
-"
-        fi
-      fi
-    fi
-
-    # Auto-install Python LSP plugin (pyright) when Python project/prompt detected
-    # v1.0.196: Uses --scope $LSP_PLUGIN_SCOPE (default: project)
-    if [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]]; then
-      PYRIGHT_INSTALLED=$(jq -r '."pyright@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      if [[ "$PYRIGHT_INSTALLED" != "true" ]] && command -v claude >/dev/null 2>&1; then
-        if timeout 15 claude plugin install pyright@claude-code-lsps --scope $LSP_PLUGIN_SCOPE >/dev/null 2>&1; then
-          LSP_INSTALL_MSG="${LSP_INSTALL_MSG}✅ **Python LSP installed**: \`pyright@claude-code-lsps\` (scope: $LSP_PLUGIN_SCOPE)
-"
-        fi
-      fi
-    fi
-
-    # Auto-install Rust LSP plugin (rust-analyzer) when Rust project/prompt detected
-    # v1.0.196: Uses --scope $LSP_PLUGIN_SCOPE (default: project)
-    if [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]]; then
-      RUST_ANALYZER_INSTALLED=$(jq -r '."rust-analyzer@claude-code-lsps" // false' "$HOME/.claude/plugins/installed_plugins.json" 2>/dev/null)
-      if [[ "$RUST_ANALYZER_INSTALLED" != "true" ]] && command -v claude >/dev/null 2>&1; then
-        if timeout 15 claude plugin install rust-analyzer@claude-code-lsps --scope $LSP_PLUGIN_SCOPE >/dev/null 2>&1; then
-          LSP_INSTALL_MSG="${LSP_INSTALL_MSG}✅ **Rust LSP installed**: \`rust-analyzer@claude-code-lsps\` (scope: $LSP_PLUGIN_SCOPE)
-"
-        fi
-      fi
-    fi
-
-    if [[ -n "$LSP_INSTALL_MSG" ]]; then
-      LSP_INSTALL_MSG="${LSP_INSTALL_MSG}
----
-
-"
-    fi
-  fi
-fi
-
-# ==============================================================================
-# VSKILL MARKETPLACE PLUGIN RECOMMENDATION (v1.0.542)
-# ==============================================================================
-VSKILL_SUGGEST_MSG=""
-
-detect_vskill_recommendations() {
-  # Testing: Keyword matching logic is tested via TypeScript mirror (llm-plugin-detector.vskill.test.ts).
-  # Bash-specific behavior (grep -qwi word boundary, case/esac, TMPDIR markers) is validated manually.
-  local prompt
-  prompt=$(echo "$1" | tr '[:upper:]' '[:lower:]')
-  local suggestions=""
-  local match_count=0
-  local max_suggestions=3
-
-  # Note: On macOS, TMPDIR is per-session (/var/folders/...) so markers auto-reset.
-  # On Linux, TMPDIR often defaults to /tmp (shared), so markers may persist across sessions.
-  # This is an accepted trade-off — worst case, suggestions appear only once per reboot on Linux.
-
-  # --- mobile ---
-  # shortKeywords (word-boundary): ios, apk
-  # keywords (substring): testflight, app store, play store, react native, react-native, expo, mobile app, flutter, xcode, cocoapods, fastlane, app-store-connect, app deployment
-  if [[ $match_count -lt $max_suggestions ]]; then
-    local mobile_match="false"
-    if echo "$prompt" | grep -qwi 'ios'; then mobile_match="true"; fi
-    if [[ "$mobile_match" == "false" ]] && echo "$prompt" | grep -qwi 'apk'; then mobile_match="true"; fi
-    if [[ "$mobile_match" == "false" ]]; then
-      case "$prompt" in
-        *testflight*|*"app store"*|*"play store"*|*"react native"*|*react-native*|*expo*|*"mobile app"*|*flutter*|*xcode*|*cocoapods*|*fastlane*|*app-store-connect*|*"app deployment"*) mobile_match="true" ;;
-      esac
-    fi
-    if [[ "$mobile_match" == "true" ]]; then
-      if ! check_plugin_in_vskill_lock "mobile"; then
-        if [[ ! -f "${TMPDIR:-/tmp}/specweave-vskill-suggested-mobile" ]]; then
-          touch "${TMPDIR:-/tmp}/specweave-vskill-suggested-mobile" 2>/dev/null
-          suggestions="${suggestions}- **mobile** (App Store, TestFlight, mobile CI/CD): \`vskill i anton-abyzov/vskill/appstore\`\n"
-          match_count=$((match_count + 1))
-        fi
-      fi
-    fi
-  fi
-
-  # --- google-workspace ---
-  # shortKeywords: (none)
-  # keywords (substring): google sheets, google docs, google drive, google slides, google calendar, google workspace, gmail api, sheets api
-  if [[ $match_count -lt $max_suggestions ]]; then
-    local gws_match="false"
-    case "$prompt" in
-      *"google sheets"*|*"google docs"*|*"google drive"*|*"google slides"*|*"google calendar"*|*"google workspace"*|*"gmail api"*|*"sheets api"*) gws_match="true" ;;
-    esac
-    if [[ "$gws_match" == "true" ]]; then
-      if ! check_plugin_in_vskill_lock "google-workspace"; then
-        if [[ ! -f "${TMPDIR:-/tmp}/specweave-vskill-suggested-google-workspace" ]]; then
-          touch "${TMPDIR:-/tmp}/specweave-vskill-suggested-google-workspace" 2>/dev/null
-          suggestions="${suggestions}- **google-workspace** (Gmail, Drive, Sheets, Docs, Calendar): \`vskill i anton-abyzov/vskill/gws\`\n"
-          match_count=$((match_count + 1))
-        fi
-      fi
-    fi
-  fi
-
-  # --- marketing ---
-  # shortKeywords: (none)
-  # keywords (substring): linkedin, instagram, social media, twitter, facebook, tiktok, content marketing, social post, blog post, newsletter, slack messaging
-  if [[ $match_count -lt $max_suggestions ]]; then
-    local mktg_match="false"
-    case "$prompt" in
-      *linkedin*|*instagram*|*"social media"*|*twitter*|*facebook*|*tiktok*|*"content marketing"*|*"social post"*|*"blog post"*|*newsletter*|*"slack messaging"*) mktg_match="true" ;;
-    esac
-    if [[ "$mktg_match" == "true" ]]; then
-      if ! check_plugin_in_vskill_lock "marketing"; then
-        if [[ ! -f "${TMPDIR:-/tmp}/specweave-vskill-suggested-marketing" ]]; then
-          touch "${TMPDIR:-/tmp}/specweave-vskill-suggested-marketing" 2>/dev/null
-          suggestions="${suggestions}- **marketing** (slack-messaging, social-media-posting): \`vskill i anton-abyzov/vskill/slack-messaging\`\n"
-          match_count=$((match_count + 1))
-        fi
-      fi
-    fi
-  fi
-
-  # --- productivity ---
-  # shortKeywords: (none)
-  # keywords (substring): notion, todoist, trello, asana, monday.com, obsidian, time tracking, project management, task management
-  if [[ $match_count -lt $max_suggestions ]]; then
-    local prod_match="false"
-    case "$prompt" in
-      *notion*|*todoist*|*trello*|*asana*|*monday.com*|*obsidian*|*"time tracking"*|*"project management"*|*"task management"*) prod_match="true" ;;
-    esac
-    if [[ "$prod_match" == "true" ]]; then
-      if ! check_plugin_in_vskill_lock "productivity"; then
-        if [[ ! -f "${TMPDIR:-/tmp}/specweave-vskill-suggested-productivity" ]]; then
-          touch "${TMPDIR:-/tmp}/specweave-vskill-suggested-productivity" 2>/dev/null
-          suggestions="${suggestions}- **productivity** (Notion, Todoist, Trello, Asana, Obsidian): \`vskill i anton-abyzov/vskill/survey-passing\`\n"
-          match_count=$((match_count + 1))
-        fi
-      fi
-    fi
-  fi
-
-  # --- skills ---
-  # shortKeywords: (none)
-  # keywords (substring): find plugin, search plugin, discover skill, browse marketplace, vskill search, plugin search
-  if [[ $match_count -lt $max_suggestions ]]; then
-    local skills_match="false"
-    case "$prompt" in
-      *"find plugin"*|*"search plugin"*|*"discover skill"*|*"browse marketplace"*|*"vskill search"*|*"plugin search"*) skills_match="true" ;;
-    esac
-    if [[ "$skills_match" == "true" ]]; then
-      if ! check_plugin_in_vskill_lock "skills"; then
-        if [[ ! -f "${TMPDIR:-/tmp}/specweave-vskill-suggested-skills" ]]; then
-          touch "${TMPDIR:-/tmp}/specweave-vskill-suggested-skills" 2>/dev/null
-          suggestions="${suggestions}- **skills** (skill discovery and installation): \`vskill i anton-abyzov/vskill/scout\`\n"
-          match_count=$((match_count + 1))
-        fi
-      fi
-    fi
-  fi
-
-  if [[ -n "$suggestions" ]]; then
-    printf '**Suggested vskill plugins** (not yet installed):\n%b' "$suggestions"
-  fi
-}
-
-if [[ -n "$PROMPT" ]]; then
-  VSKILL_SUGGEST_MSG=$(detect_vskill_recommendations "$PROMPT")
-fi
-
-# ==============================================================================
-# EXPLICIT LSP REQUEST DETECTION (v1.0.181)
-# ==============================================================================
-# Detects when users explicitly ask to "use LSP" for tasks like "find references"
-# and explains that LSP in Claude Code provides background enhancement, not explicit tools.
-# v1.0.201: SEMANTIC LSP VIA CLI FALLBACK
-# When user asks for LSP features (find references, go to definition, etc.),
-# instruct Claude to use `specweave lsp` commands which provide REAL semantic analysis
-# via TsServerClient - NOT grep text matching!
-# ==============================================================================
-# v1.0.235: LSP AUTO-INSTALL + CLI INSTRUCTIONS
-# ==============================================================================
-# When user requests LSP operations (find references, go to definition, etc.):
-# 1. Detect which language from prompt/project context
-# 2. Check if that language's LSP binary + plugin are installed
-# 3. If missing, return AUTO-INSTALL commands for Claude to run silently
-# 4. Then provide CLI instructions for the actual LSP operation
-LSP_EXPLICIT_REQUEST_MSG=""
-if echo "$PROMPT" | grep -qiE "(use|with|via)[[:space:]]+(the[[:space:]]+)?LSP|LSP[[:space:]]+(find|get|show|goto|hover|definition|references|implementations)|(find|get|show)[[:space:]]+(all[[:space:]]+)?references|go[[:space:]]?to[[:space:]]?definition|where[[:space:]]+(is|are).*defined|what[[:space:]]+(uses|calls)|who[[:space:]]+(uses|calls)"; then
-
-  # Build auto-install commands for missing LSP servers
-  LSP_AUTO_INSTALL_CMDS=""
-  _lsp_check_and_install() {
-    local lang="$1" binary="$2" binary_install="$3" plugin="$4" marketplace="${LSP_MARKETPLACE:-claude-code-lsps}"
-    # Check if binary is installed
-    if ! command -v "$binary" >/dev/null 2>&1; then
-      LSP_AUTO_INSTALL_CMDS="${LSP_AUTO_INSTALL_CMDS}
-- **${lang}** binary missing: \`${binary_install}\`"
-    fi
-    # Check if plugin is installed
-    local plugins_file="$HOME/.claude/plugins/installed_plugins.json"
-    if [[ -f "$plugins_file" ]]; then
-      if ! grep -q "\"${plugin}@" "$plugins_file" 2>/dev/null; then
-        LSP_AUTO_INSTALL_CMDS="${LSP_AUTO_INSTALL_CMDS}
-- **${lang}** plugin missing: \`claude plugin install ${plugin}@${marketplace} --scope ${LSP_PLUGIN_SCOPE:-project}\`"
-      fi
-    else
-      LSP_AUTO_INSTALL_CMDS="${LSP_AUTO_INSTALL_CMDS}
-- **${lang}** plugin missing: \`claude plugin install ${plugin}@${marketplace} --scope ${LSP_PLUGIN_SCOPE:-project}\`"
-    fi
-  }
-
-  # Check each detected language
-  if [[ "$LSP_PROJECT_NEEDS_TS" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_TS" == "true" ]]; then
-    _lsp_check_and_install "TypeScript" "typescript-language-server" "npm install -g typescript-language-server typescript" "vtsls"
-  fi
-  if [[ "$LSP_PROJECT_NEEDS_PY" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_PY" == "true" ]]; then
-    _lsp_check_and_install "Python" "pyright-langserver" "pip install pyright" "pyright"
-  fi
-  if [[ "$LSP_PROJECT_NEEDS_RUST" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_RUST" == "true" ]]; then
-    _lsp_check_and_install "Rust" "rust-analyzer" "rustup component add rust-analyzer" "rust-analyzer"
-  fi
-  if [[ "$LSP_PROJECT_NEEDS_CSHARP" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_CSHARP" == "true" ]]; then
-    _lsp_check_and_install "C#" "csharp-ls" "dotnet tool install -g csharp-ls" "csharp-lsp"
-  fi
-  if [[ "$LSP_PROJECT_NEEDS_GO" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_GO" == "true" ]]; then
-    _lsp_check_and_install "Go" "gopls" "go install golang.org/x/tools/gopls@latest" "gopls"
-  fi
-  if [[ "$LSP_PROJECT_NEEDS_JAVA" == "true" ]] || [[ "$LSP_PROMPT_NEEDS_JAVA" == "true" ]]; then
-    _lsp_check_and_install "Java" "jdtls" "brew install jdtls" "jdtls"
-  fi
-
-  # Build the message
-  if [[ -n "$LSP_AUTO_INSTALL_CMDS" ]]; then
-    LSP_EXPLICIT_REQUEST_MSG="⚡ **LSP AUTO-INSTALL REQUIRED**
-
-The following LSP components are missing for this request. **Install them NOW before proceeding:**
-${LSP_AUTO_INSTALL_CMDS}
-
-**MANDATORY**: Run ALL missing install commands above using Bash tool, then use \`specweave lsp\` CLI:
-
-\`\`\`bash
-specweave lsp refs <file> <symbol>     # Find semantic references
-specweave lsp def <file> <symbol>      # Go to definition
-specweave lsp hover <file> <symbol>    # Type information
-specweave lsp symbols <file>           # List symbols
-\`\`\`
-
-⚠️ Do NOT use Grep for \"find references\" - use LSP for semantic accuracy.
-
----
-
-"
-  else
-    LSP_EXPLICIT_REQUEST_MSG="🎯 **LSP Semantic Analysis Available**
-
-Use SpecWeave LSP CLI for **semantic code intelligence** (not grep text matching):
-
-\`\`\`bash
-specweave lsp refs <file> <symbol>     # Find semantic references
-specweave lsp def <file> <symbol>      # Go to definition
-specweave lsp hover <file> <symbol>    # Type information
-specweave lsp symbols <file>           # List symbols
-specweave lsp search <query>           # Workspace symbol search
-\`\`\`
-
-⚠️ Do NOT use Grep for \"find references\" - use LSP for semantic accuracy.
-
----
-
-"
-  fi
-fi
-
-# Only run if features are enabled and not disabled via env
-if [[ "${SPECWEAVE_DISABLE_AUTO_LOAD:-0}" != "1" ]] && [[ "${SPECWEAVE_DISABLE_HOOKS:-0}" != "1" ]]; then
-  if [[ "$PLUGIN_AUTOLOAD_ENABLED" == "true" ]] || [[ "$INCREMENT_ASSIST_ENABLED" == "true" ]]; then
-
-    # Quick skip: already using /sw: commands (user is in workflow)
-    if ! echo "$PROMPT" | grep -qE "^[[:space:]]*/sw:"; then
-
-      # BYPASS: Native Claude Code slash commands (e.g., /context, /help, /doctor)
-      # Prevents 15s detect-intent timeout → LLM_DETECTION_FAILED → keyword fallback
-      # that falsely matches "test" as substring inside "/context". Pattern matches
-      # /word or /word-word prompts that don't mention specweave.
-      if echo "$PROMPT" | grep -qE "^[[:space:]]*/[a-z][a-z0-9-]*([[:space:]]|$)" && \
-         ! echo "$PROMPT" | grep -qiE "specweave"; then
-        echo '{"decision":"approve"}'
-        exit 0
-      fi
-
-      # Check if specweave CLI is available
-      if command -v specweave >/dev/null 2>&1; then
-        # Setup logging (use project root, never create dirs at $HOME)
-        if [[ -n "$SW_PROJECT_ROOT" ]]; then
-          LAZY_LOAD_LOG="$SW_PROJECT_ROOT/.specweave/logs/lazy-loading.log"
-        else
-          LAZY_LOAD_LOG="/dev/null"
-        fi
-
-        # Per-session cache to avoid redundant LLM calls (30 min TTL)
-        if [[ -n "$SW_PROJECT_ROOT" ]]; then
-          PROMPT_CACHE_DIR="$SW_PROJECT_ROOT/.specweave/state/prompt-cache"
-          mkdir -p "$PROMPT_CACHE_DIR" 2>/dev/null
-        else
-          PROMPT_CACHE_DIR="${TMPDIR:-/tmp}/specweave-prompt-cache"
-          mkdir -p "$PROMPT_CACHE_DIR" 2>/dev/null
-        fi
-        PROMPT_HASH=$(echo "$PROMPT" | md5sum 2>/dev/null | cut -c1-16 || md5 -qs "$PROMPT" 2>/dev/null | cut -c1-16 || echo "nohash")
-        CACHE_FILE="$PROMPT_CACHE_DIR/${PROMPT_HASH}.json"
-
-        DETECT_OUTPUT=""
-        SHOULD_CALL_LLM=true
-
-        # Check cache
-        if [[ -f "$CACHE_FILE" ]]; then
-          CACHE_AGE=$(($(date +%s) - $(stat -f%m "$CACHE_FILE" 2>/dev/null || stat -c%Y "$CACHE_FILE" 2>/dev/null || echo 0)))
-          if [[ "$CACHE_AGE" -lt 1800 ]]; then
-            DETECT_OUTPUT=$(cat "$CACHE_FILE" 2>/dev/null)
-            SHOULD_CALL_LLM=false
-            echo "[$(date -Iseconds)] detect-intent | cached=true | age=${CACHE_AGE}s" >> "$LAZY_LOAD_LOG"
-          fi
-        fi
-
-        # Call LLM if not cached
-        if [[ "$SHOULD_CALL_LLM" == "true" ]]; then
-          START_TIME=$(perl -MTime::HiRes=time -e 'printf "%.0f", time*1000' 2>/dev/null || echo $(($(date +%s) * 1000)))
-
-          # Write prompt to temp file to avoid all escaping issues (v1.0.153)
-          PROMPT_TMP_FILE=$(mktemp 2>/dev/null || echo "/tmp/specweave-prompt-$$")
-          printf '%s' "$PROMPT" > "$PROMPT_TMP_FILE"
-
-          # ONE LLM call for BOTH plugins and increment (using --file flag)
-          if command -v timeout >/dev/null 2>&1; then
-            # v1.0.159: Reduced timeout to 15s with --setting-sources "" optimization
-            DETECT_OUTPUT=$(timeout 15 specweave detect-intent --file "$PROMPT_TMP_FILE" 2>/dev/null)
-          else
-            DETECT_OUTPUT=$(specweave detect-intent --file "$PROMPT_TMP_FILE" 2>/dev/null)
-          fi
-
-          # Clean up temp file
-          rm -f "$PROMPT_TMP_FILE" 2>/dev/null
-
-          END_TIME=$(perl -MTime::HiRes=time -e 'printf "%.0f", time*1000' 2>/dev/null || echo $(($(date +%s) * 1000)))
-          DURATION=$((END_TIME - START_TIME))
-
-          # Cache result
-          [[ -n "$DETECT_OUTPUT" ]] && echo "$DETECT_OUTPUT" > "$CACHE_FILE" 2>/dev/null
-          echo "[$(date -Iseconds)] detect-intent | duration=${DURATION}ms | cached=false" >> "$LAZY_LOAD_LOG"
-        fi
-
-        # Parse JSON response (extract complete JSON object from multi-line output)
-        LLM_DETECTION_FAILED=false
-        if [[ -n "$DETECT_OUTPUT" ]] && command -v jq >/dev/null 2>&1; then
-          # Extract JSON using awk: from first { to last } (handles multi-line JSON)
-          JSON_OUTPUT=$(echo "$DETECT_OUTPUT" | awk '/^\{/{found=1} found{print} /^\}/{if(found) exit}')
-
-          if [[ -n "$JSON_OUTPUT" ]]; then
-            # ==================================================================
-            # ERROR RESPONSE DETECTION (v1.0.337)
-            # ==================================================================
-            # When detect-intent returns valid JSON but with error data (e.g.,
-            # nested session error, timeout, auth failure), the JSON has
-            # installMessage with error text, confidence=0, and no increment
-            # field. Without this check, the hook treats error responses as
-            # "LLM worked, nothing needed" and skips keyword fallback entirely.
-            INSTALL_MSG=$(echo "$JSON_OUTPUT" | jq -r '.installMessage // empty' 2>/dev/null)
-            HAS_INCREMENT=$(echo "$JSON_OUTPUT" | jq -r 'has("increment")' 2>/dev/null)
-            RESP_CONFIDENCE=$(echo "$JSON_OUTPUT" | jq -r '.confidence // 0' 2>/dev/null)
-            if [[ -n "$INSTALL_MSG" && "$HAS_INCREMENT" != "true" && "$RESP_CONFIDENCE" == "0" ]]; then
-              LLM_DETECTION_FAILED=true
-              echo "[$(date -Iseconds)] LLM detection failed | reason=error_response | msg=${INSTALL_MSG:0:100}" >> "$LAZY_LOAD_LOG"
-            fi
-
-            # ==================================================================
-            # ON-DEMAND PLUGIN INSTALL (v1.0.540 — restored)
-            # ==================================================================
-            # When LLM detect-intent identifies plugins needed for this prompt,
-            # install any that aren't already present. Uses specweave CLI's
-            # --plugin flag for targeted install. Session markers prevent
-            # duplicate installs within the same Claude Code session.
-            # ==================================================================
-            DETECTED_PLUGINS=$(echo "$JSON_OUTPUT" | jq -r '.plugins[]? // empty' 2>/dev/null)
-            if [[ -n "$DETECTED_PLUGINS" ]]; then
-              CACHE_BASE="${HOME}/.claude/plugins/cache/specweave"
-              SKILLS_BASE="${SW_PROJECT_ROOT:-.}/.claude/skills"
-              # Use PPID (parent process = Claude Code) for session-level uniqueness
-              SESSION_MARKER_DIR="${TMPDIR:-/tmp}/specweave-ondemand-${PPID:-$$}"
-              mkdir -p "$SESSION_MARKER_DIR" 2>/dev/null
-              ONDEMAND_PIDS=()
-
-              while IFS= read -r PLUGIN_NAME; do
-                [[ -z "$PLUGIN_NAME" ]] && continue
-                [[ "$PLUGIN_NAME" == "sw" ]] && continue  # core always installed
-                # Sanitize: only allow alphanumeric, dash, underscore (prevent injection)
-                if [[ ! "$PLUGIN_NAME" =~ ^[a-zA-Z0-9_-]+$ ]]; then
-                  continue
-                fi
-
-                # Idempotency: skip if already installed or already attempted this session
-                if [[ -d "$CACHE_BASE/$PLUGIN_NAME" ]] || \
-                   [[ -d "$SKILLS_BASE/$PLUGIN_NAME" ]] || \
-                   [[ -f "$SESSION_MARKER_DIR/$PLUGIN_NAME" ]]; then
-                  continue
-                fi
-
-                # Install via CLI (handles both native and direct-copy modes)
-                specweave refresh-plugins --plugin "$PLUGIN_NAME" --quiet 2>/dev/null &
-                ONDEMAND_PIDS+=($!)
-                touch "$SESSION_MARKER_DIR/$PLUGIN_NAME" 2>/dev/null
-
-                AUTOLOAD_PLUGINS_MSG="${AUTOLOAD_PLUGINS_MSG}Installed plugin: ${PLUGIN_NAME} (on-demand)."$'\n'
-              done <<< "$DETECTED_PLUGINS"
-
-              # Wait for all background installs with 5s timeout
-              if [[ ${#ONDEMAND_PIDS[@]} -gt 0 ]]; then
-                for pid in "${ONDEMAND_PIDS[@]}"; do
-                  ( sleep 5 && kill "$pid" 2>/dev/null ) &
-                  local timeout_pid=$!
-                  wait "$pid" 2>/dev/null
-                  kill "$timeout_pid" 2>/dev/null 2>&1
-                done
-              fi
-            fi
-
-            # ==================================================================
-            # EXTRACT ROUTING INFO EARLY (v1.0.155 - needed for agent directives)
-            # ==================================================================
-            ROUTING_SKILLS_COUNT=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills // [] | length' 2>/dev/null)
-
-            # ==================================================================
-            # INCREMENT SUGGESTION (from LLM response)
-            # ==================================================================
-            if [[ "$INCREMENT_ASSIST_ENABLED" == "true" ]]; then
-              INC_ACTION=$(echo "$JSON_OUTPUT" | jq -r '.increment.action // "none"' 2>/dev/null)
-              INC_CONF=$(echo "$JSON_OUTPUT" | jq -r '.increment.confidence // 0' 2>/dev/null)
-              INC_NAME=$(echo "$JSON_OUTPUT" | jq -r '.increment.suggestedName // empty' 2>/dev/null)
-              INC_REASON=$(echo "$JSON_OUTPUT" | jq -r '.increment.reasoning // empty' 2>/dev/null)
-              INC_KEYWORD=$(echo "$JSON_OUTPUT" | jq -r '.increment.relatedKeyword // empty' 2>/dev/null)
-              # v1.0.168: LLM decides if mandatory (not config-based)
-              INC_MANDATORY=$(echo "$JSON_OUTPUT" | jq -r '.increment.mandatory // false' 2>/dev/null)
-
-              # Config-based override: if incrementAssist.mandatory=true in config,
-              # force mandatory for ALL detected implementation work (action != "none")
-              if [[ "$INCREMENT_MANDATORY_CONFIG" == "true" && "$INC_ACTION" != "none" ]]; then
-                INC_MANDATORY="true"
-              fi
-
-              # v1.0.168: Parse skill invocation recommendation
-              SKILL_INVOCATION=$(echo "$JSON_OUTPUT" | jq -r '.skillInvocation.skill // empty' 2>/dev/null)
-              SKILL_REASON=$(echo "$JSON_OUTPUT" | jq -r '.skillInvocation.reason // empty' 2>/dev/null)
-              SKILL_MANDATORY=$(echo "$JSON_OUTPUT" | jq -r '.skillInvocation.mandatory // false' 2>/dev/null)
-
-              # v1.0.198: Parse LSP recommendation from unified LLM detection
-              LSP_LLM_NEEDED=$(echo "$JSON_OUTPUT" | jq -r '.lsp.needed // false' 2>/dev/null)
-              LSP_LLM_OPERATION=$(echo "$JSON_OUTPUT" | jq -r '.lsp.operation // empty' 2>/dev/null)
-              LSP_LLM_LANGUAGE=$(echo "$JSON_OUTPUT" | jq -r '.lsp.language // empty' 2>/dev/null)
-              LSP_LLM_WARMUP=$(echo "$JSON_OUTPUT" | jq -r '.lsp.warmupRequired // false' 2>/dev/null)
-
-              # Override grep-based detection with LLM decision
-              if [[ "$LSP_LLM_NEEDED" == "true" ]]; then
-                LSP_REQUEST_DETECTED="true"
-                # v1.0.235: Use LLM language detection for all supported languages
-                case "$LSP_LLM_LANGUAGE" in
-                  typescript|javascript) LSP_PROMPT_NEEDS_TS="true" ;;
-                  python) LSP_PROMPT_NEEDS_PY="true" ;;
-                  rust) LSP_PROMPT_NEEDS_RUST="true" ;;
-                  csharp|c#) LSP_PROMPT_NEEDS_CSHARP="true" ;;
-                  go|golang) LSP_PROMPT_NEEDS_GO="true" ;;
-                  java|kotlin) LSP_PROMPT_NEEDS_JAVA="true" ;;
-                esac
-              fi
-
-              # Check confidence threshold
-              ABOVE=$(echo "$INC_CONF >= $INCREMENT_CONFIDENCE_THRESHOLD" | bc -l 2>/dev/null || echo 0)
-
-              if [[ "$ABOVE" == "1" ]]; then
-                AUTOLOAD_PREFIX=""
-                # v1.0.166: Prepend external folder warning if detected
-                [[ -n "$EXTERNAL_FOLDER_DETECTED" ]] && AUTOLOAD_PREFIX="${EXTERNAL_FOLDER_DETECTED}"
-                # v1.0.179: Prepend LSP warning if language servers missing
-                [[ -n "$LSP_WARNING_MSG" ]] && AUTOLOAD_PREFIX="${AUTOLOAD_PREFIX}${LSP_WARNING_MSG}"
-                # v1.0.180: Prepend explicit LSP request explanation
-                [[ -n "$LSP_EXPLICIT_REQUEST_MSG" ]] && AUTOLOAD_PREFIX="${AUTOLOAD_PREFIX}${LSP_EXPLICIT_REQUEST_MSG}"
-                # v1.0.542: Prepend vskill plugin suggestions
-                [[ -n "$VSKILL_SUGGEST_MSG" ]] && AUTOLOAD_PREFIX="${AUTOLOAD_PREFIX}${VSKILL_SUGGEST_MSG}
-"
-                [[ -n "$AUTOLOAD_PLUGINS_MSG" ]] && AUTOLOAD_PREFIX="${AUTOLOAD_PREFIX}${AUTOLOAD_PLUGINS_MSG}
-
-"
-                # Build agent spawn directive if routing skills available (v1.0.155)
-                AGENT_DIRECTIVE=""
-                # v1.0.168: Skill invocation directive (takes precedence over routing)
-                # Skill memories now loaded via DCI in SKILL.md (no hook injection)
-                # v1.0.260: Compacted AGENT_DIRECTIVE to save context budget
-                if [[ -n "$SKILL_INVOCATION" ]]; then
-                  if [[ "$SKILL_MANDATORY" == "true" ]]; then
-                    AGENT_DIRECTIVE="
-MANDATORY: Also call \`Skill({ skill: \"${SKILL_INVOCATION}\" })\` — ${SKILL_REASON:-specialized support needed}."
-                  else
-                    AGENT_DIRECTIVE="
-Recommended: \`Skill({ skill: \"${SKILL_INVOCATION}\" })\` — ${SKILL_REASON:-specialized support for this task}."
-                  fi
-                elif [[ "$ROUTING_SKILLS_COUNT" -gt 0 ]]; then
-                  PRIMARY_PLUGIN=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .plugin // empty' 2>/dev/null | head -1)
-                  PRIMARY_SKILL_NAME=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .name // empty' 2>/dev/null | head -1)
-                  PRIMARY_REASON=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .reason // empty' 2>/dev/null | head -1)
-                  if [[ -n "$PRIMARY_PLUGIN" && -n "$PRIMARY_SKILL_NAME" ]]; then
-                    # v1.0.260: Compacted routing directive to save context budget
-                    AGENT_DIRECTIVE="
-Then spawn agent: \`Task({ subagent_type: \"${PRIMARY_PLUGIN}:${PRIMARY_SKILL_NAME}\", description: \"${PRIMARY_REASON:-Implementation}\" })\`"
-                  fi
-                fi
-
-                case "$INC_ACTION" in
-                  new)
-                    CMD="/sw:increment"
-                    [[ -n "$INC_NAME" ]] && CMD="/sw:increment \"$INC_NAME\""
-
-                    # v1.0.169+: Call sw:increment skill directly
-                    # Pass user prompt so skill can extract context
-                    # INC_MANDATORY comes from detect-intent LLM response
-                    if [[ "$INC_MANDATORY" == "true" ]]; then
-                      # v1.0.260: Removed prompt embedding from SKILL FIRST args to save ~800 chars.
-                      # The skill reads the user's prompt from conversation context (it's already there).
-
-                      # v1.0.243: Smart interview gate — LLM assesses prompt completeness
-                      DEEP_INTERVIEW_MSG=""
-                      if [[ "$DEEP_INTERVIEW_ENABLED" == "true" ]]; then
-                        DEEP_INTERVIEW_MSG=" Assess prompt completeness first — sw:pm skill will assess complexity and determine question count. If gaps exist, ask targeted questions (count depends on complexity) before calling."
-                      fi
-
-                      MSG="${WIP_WARNING}${AUTOLOAD_PREFIX}SKILL FIRST: \`Skill({ skill: \"sw:increment\" })\` — call BEFORE implementation.
-Detection: ${INC_REASON} (confidence: ${INC_CONF}).${AGENT_DIRECTIVE}${DEEP_INTERVIEW_MSG}
-After increment, chain domain skills per tech stack (see CLAUDE.md Skill Chaining)."
-                      output_approve_with_context "$MSG"
-                      exit 0
-                    else
-                      # v1.0.260: Removed prompt embedding to save context budget
-                      MSG="${WIP_WARNING}${AUTOLOAD_PREFIX}Increment suggested: \`Skill({ skill: \"sw:increment\" })\` or \`$CMD\`. Reason: $INC_REASON${AGENT_DIRECTIVE}"
-                      output_approve_with_context "$MSG"
-                      exit 0
-                    fi
-                    ;;
-
-                  hotfix)
-                    # v1.0.260: Removed prompt embedding to save context budget
-                    MSG="${WIP_WARNING}${AUTOLOAD_PREFIX}Hotfix detected: \`Skill({ skill: \"sw:increment\", args: \"--type=hotfix\" })\`. Reason: $INC_REASON"
-                    output_approve_with_context "$MSG"
-                    exit 0
-                    ;;
-
-                  reopen)
-                    HINT=""
-                    [[ -n "$INC_KEYWORD" ]] && HINT=" (look for: *$INC_KEYWORD*)"
-                    MSG="${AUTOLOAD_PREFIX}Related to previous work$HINT. Consider: \`/sw:status\` then \`specweave resume <id>\`. Reason: $INC_REASON"
-                    output_approve_with_context "$MSG"
-                    exit 0
-                    ;;
-
-                  small_fix)
-                    # v1.0.260: Removed prompt embedding to save context budget
-                    CMD_SMALLFIX="/sw:increment"
-                    [[ -n "$INC_NAME" ]] && CMD_SMALLFIX="/sw:increment \"$INC_NAME\""
-
-                    MSG="${WIP_WARNING}${AUTOLOAD_PREFIX}Small change — consider tracking: \`Skill({ skill: \"sw:increment\" })\` or \`$CMD_SMALLFIX\`. Reason: $INC_REASON${AGENT_DIRECTIVE}"
-                    output_approve_with_context "$MSG"
-                    exit 0
-                    ;;
-                esac
-              fi
-
-              # ==================================================================
-              # SKILL-ONLY INVOCATION (v1.0.196)
-              # Handle skill invocation when NO increment is suggested but LLM
-              # recommends a skill (e.g., LSP skill for "find references" prompts)
-              # ==================================================================
-              if [[ "$ABOVE" != "1" && -n "$SKILL_INVOCATION" ]]; then
-                # Build prefix messages
-                SKILL_ONLY_PREFIX=""
-                [[ -n "$LSP_WARNING_MSG" ]] && SKILL_ONLY_PREFIX="${LSP_WARNING_MSG}"
-                [[ -n "$LSP_EXPLICIT_REQUEST_MSG" ]] && SKILL_ONLY_PREFIX="${SKILL_ONLY_PREFIX}${LSP_EXPLICIT_REQUEST_MSG}"
-
-                if [[ "$SKILL_MANDATORY" == "true" ]]; then
-                  MSG="${SKILL_ONLY_PREFIX}SKILL REQUIRED: \`Skill({ skill: \"${SKILL_INVOCATION}\" })\` — call before proceeding. ${SKILL_REASON:-Specialized support needed.}"
-                  output_approve_with_context "$MSG"
-                  exit 0
-                else
-                  MSG="${SKILL_ONLY_PREFIX}Skill recommended: \`Skill({ skill: \"${SKILL_INVOCATION}\" })\`. ${SKILL_REASON:-Specialized support for this task.}"
-                  output_approve_with_context "$MSG"
-                  exit 0
-                fi
-              fi
-            fi
-
-            # ==================================================================
-            # SKILL ROUTING (from LLM response) - v1.0.150+
-            # ==================================================================
-            # Extract skill routing for brain message
-            ROUTING_SKILLS_COUNT=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills // [] | length' 2>/dev/null)
-            ROUTING_MSG=""
-
-            if [[ "$ROUTING_SKILLS_COUNT" -gt 0 ]]; then
-              # Extract primary skill
-              PRIMARY_SKILL=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .fullName // empty' 2>/dev/null | head -1)
-              PRIMARY_INVOKE=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .invokeWhen // "after_increment"' 2>/dev/null | head -1)
-              PRIMARY_REASON=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .reason // empty' 2>/dev/null | head -1)
-
-              # Extract secondary skills
-              SECONDARY_SKILLS=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "secondary") | .fullName' 2>/dev/null | tr '\n' ', ' | sed 's/,$//')
-
-              # Extract workflow info
-              SUGGEST_PLAN=$(echo "$JSON_OUTPUT" | jq -r '.routing.workflow.suggestPlanMode // false' 2>/dev/null)
-              WORKFLOW_PHASES=$(echo "$JSON_OUTPUT" | jq -r '.routing.workflow.phases[]?' 2>/dev/null | tr '\n' ' ')
-              ROUTING_REASON=$(echo "$JSON_OUTPUT" | jq -r '.routing.reasoning // empty' 2>/dev/null)
-
-              echo "[$(date -Iseconds)] routing | primary=${PRIMARY_SKILL:-none} | secondary=${SECONDARY_SKILLS:-none} | invokeWhen=${PRIMARY_INVOKE}" >> "$LAZY_LOAD_LOG"
-            fi
-
-            # ==================================================================
-            # BUILD BRAIN MESSAGE (combining all analysis)
-            # ==================================================================
-            # Only show brain message if we have meaningful routing info
-            if [[ "$ROUTING_SKILLS_COUNT" -gt 0 || -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-              BRAIN_MSG=""
-
-              # Compact router output — one line per decision
-              if [[ "$INC_ACTION" == "new" || "$INC_ACTION" == "hotfix" ]]; then
-                BRAIN_MSG+="Increment: create \\\"${INC_NAME:-new-feature}\\\" (${INC_ACTION}). "
-              elif [[ "$INC_ACTION" == "reopen" ]]; then
-                BRAIN_MSG+="Increment: reopen existing"
-                [[ -n "$INC_KEYWORD" ]] && BRAIN_MSG+=" (${INC_KEYWORD})"
-                BRAIN_MSG+=". "
-              fi
-
-              if [[ -n "$PRIMARY_SKILL" ]]; then
-                # Extract agent type for Task tool
-                PRIMARY_PLUGIN=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .plugin // empty' 2>/dev/null | head -1)
-                PRIMARY_SKILL_NAME=$(echo "$JSON_OUTPUT" | jq -r '.routing.skills[] | select(.priority == "primary") | .name // empty' 2>/dev/null | head -1)
-                BRAIN_MSG+="Primary skill: ${PRIMARY_SKILL}"
-                # v2.1.0: Use plugin:skill format for agent type (e.g., backend:dotnet)
-                [[ -n "$PRIMARY_PLUGIN" && -n "$PRIMARY_SKILL_NAME" ]] && BRAIN_MSG+=" (agent: ${PRIMARY_PLUGIN}:${PRIMARY_SKILL_NAME})"
-                BRAIN_MSG+=", invoke: ${PRIMARY_INVOKE:-after_increment}. "
-                [[ -n "$SECONDARY_SKILLS" ]] && BRAIN_MSG+="Also: ${SECONDARY_SKILLS}. "
-              fi
-
-              [[ "$SUGGEST_PLAN" == "true" ]] && BRAIN_MSG+="Suggest plan mode. "
-
-              output_approve_with_context "$BRAIN_MSG"
-              exit 0
-            fi
-
-            # If plugins loaded but no routing, show just plugins (fallback)
-            if [[ -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-              output_approve_with_context "$AUTOLOAD_PLUGINS_MSG"
-              exit 0
-            fi
-          else
-            # JSON_OUTPUT was empty - LLM detection failed
-            LLM_DETECTION_FAILED=true
-            echo "[$(date -Iseconds)] LLM detection failed | reason=empty_json_output" >> "$LAZY_LOAD_LOG"
-          fi
-        else
-          # DETECT_OUTPUT was empty or jq not available
-          LLM_DETECTION_FAILED=true
-          echo "[$(date -Iseconds)] LLM detection failed | reason=empty_detect_output_or_no_jq" >> "$LAZY_LOAD_LOG"
-        fi
-
-        # ==================================================================
-        # KEYWORD FALLBACK: PLUGIN INSTALLATION (v1.0.540 — note)
-        # ==================================================================
-        # On-demand plugin installation restored in v1.0.540 via the LLM
-        # detection path above. Keyword-based fallback is not needed here
-        # because the LLM detect-intent already identifies needed plugins.
-        # ==================================================================
-
-        # ==================================================================
-        # KEYWORD FALLBACK FOR INCREMENT DISCIPLINE (v1.0.257)
-        # ==================================================================
-        # When LLM detection fails/times out, use keyword matching for
-        # increment suggestions.
-        if [[ "$LLM_DETECTION_FAILED" == "true" && "$INCREMENT_ASSIST_ENABLED" == "true" ]]; then
-          # Check for implementation-intent keywords
-          # v1.0.261: Expanded from 20 to 65+ keywords across 9 categories:
-          # Original, Investigation, Analysis, Problem-solving, Optimization,
-          # Security, Documentation, DevOps/Data, Structural
-          if echo "$PROMPT" | grep -qiE "(test|component|feature|fix|refactor|setup|configure|integrate|migrate|upgrade|write|style|design|add|create|implement|build|develop|deploy|scaffold|generate|investigate|debug|troubleshoot|diagnose|trace|profile|examine|inspect|reproduce|replicate|analyze|assess|audit|evaluate|benchmark|measure|validate|solve|resolve|address|tackle|determine|optimize|improve|reduce|minimize|eliminate|simplify|streamline|secure|harden|patch|sanitize|encrypt|document|provision|containerize|dockerize|seed|populate|import|export|transform|sync|batch|remove|delete|replace|convert|extract|merge|split|wrap|unwrap|decouple|modularize)"; then
-            # Exclude PURE questions but NOT investigation/work prompts
-            # v1.0.261: "why" and "how" removed — they almost always imply work intent
-            # ("why does X fail" = investigation, "how do I fix X" = work)
-            # Patterns made more specific to avoid false negatives
-            if ! echo "$PROMPT" | grep -qiE "^[[:space:]]*(what is|what are|explain|tell me about|can you explain|does .* support|should I use|is there a|where is|when did|which one)" && \
-               ! echo "$PROMPT" | grep -qE "\?[[:space:]]*$"; then
-              # v1.0.260: Removed prompt embedding to save context budget
-              if [[ "$INCREMENT_MANDATORY_CONFIG" == "true" ]]; then
-                FALLBACK_MSG="SKILL FIRST: \`Skill({ skill: \"sw:increment\" })\` — call BEFORE implementation.
-Detection: Implementation keywords detected (LLM unavailable, keyword fallback).
-After increment, chain domain skills per tech stack (see CLAUDE.md Skill Chaining)."
-              else
-                FALLBACK_MSG="Increment suggested: \`Skill({ skill: \"sw:increment\" })\`. Reason: Implementation keywords detected (LLM unavailable, keyword fallback)."
-              fi
-              # v1.0.542: Prepend vskill suggestions to fallback message
-              [[ -n "$VSKILL_SUGGEST_MSG" ]] && FALLBACK_MSG="${VSKILL_SUGGEST_MSG}
-${FALLBACK_MSG}"
-              echo "[$(date -Iseconds)] keyword-fallback | prompt_keywords_matched=true | mandatory=$INCREMENT_MANDATORY_CONFIG" >> "$LAZY_LOAD_LOG"
-              output_approve_with_context "$FALLBACK_MSG"
-              exit 0
-            fi
-          fi
-
-          # ==================================================================
-          # ERROR-STATE DETECTION: symptom-based prompts (v1.0.261)
-          # ==================================================================
-          # Catches prompts describing failure states without action verbs:
-          # "the dashboard is broken", "login keeps failing", "app crashes on mobile"
-          # Only runs when LLM failed AND primary keyword regex didn't match.
-          if echo "$PROMPT" | grep -qiE "(is broken|keeps? failing|crash(es|ing)|hang(s|ing)|times? out|is slow|memory leak|performance issue|not working|throwing error|exception|stack trace|segfault|deadlock|race condition)"; then
-            # Require 3+ words for context (not just bare "is broken")
-            WORD_COUNT=$(echo "$PROMPT" | wc -w | tr -d ' ')
-            if [[ "$WORD_COUNT" -ge 3 ]]; then
-              if [[ "$INCREMENT_MANDATORY_CONFIG" == "true" ]]; then
-                FALLBACK_MSG="SKILL FIRST: \`Skill({ skill: \"sw:increment\" })\` — call BEFORE implementation.
-Detection: Error/failure state detected (LLM unavailable, symptom fallback).
-After increment, chain domain skills per tech stack (see CLAUDE.md Skill Chaining)."
-              else
-                FALLBACK_MSG="Increment suggested: \`Skill({ skill: \"sw:increment\" })\`. Reason: Error/failure state detected (LLM unavailable, symptom fallback)."
-              fi
-              echo "[$(date -Iseconds)] symptom-fallback | prompt_symptom_matched=true | mandatory=$INCREMENT_MANDATORY_CONFIG" >> "$LAZY_LOAD_LOG"
-              output_approve_with_context "$FALLBACK_MSG"
-              exit 0
-            fi
-          fi
-        fi
-      fi
-    fi
-  fi
-fi
-
-# ==============================================================================
-# TDD MODE CONTEXT INJECTION (v1.0.148) - Ensure Claude ALWAYS knows TDD status
-# ==============================================================================
-# Priority (highest to lowest):
-#   1. Command flag: --tdd or --strict in prompt
-#   2. Increment metadata: .specweave/increments/<active>/metadata.json
-#   3. Global config: .specweave/config.json
-#
-# When TDD is enabled, injects context into systemMessage so Claude:
-#   - Always knows to follow RED→GREEN→REFACTOR discipline
-#   - Uses /sw:tdd-cycle for guided workflow
-#   - Blocks or warns on out-of-order task completion (based on enforcement)
-
-TDD_MODE="off"
-TDD_ENFORCEMENT="warn"
-TDD_SOURCE=""
-TDD_MSG=""
-
-# Only check TDD if we're in a SpecWeave project (use resolved project root)
-if [[ -n "$SW_PROJECT_ROOT" ]] && [[ -f "$SW_PROJECT_ROOT/.specweave/config.json" ]]; then
-
-  # Step 1: Check global config (LOWEST priority)
-  if [[ -f "$SW_PROJECT_ROOT/.specweave/config.json" ]] && command -v jq >/dev/null 2>&1; then
-    GLOBAL_TDD=$(jq -r '.testing.defaultTestMode // "test-after"' "$SW_PROJECT_ROOT/.specweave/config.json" 2>/dev/null)
-    GLOBAL_ENFORCEMENT=$(jq -r '.testing.tddEnforcement // "warn"' "$SW_PROJECT_ROOT/.specweave/config.json" 2>/dev/null)
-    if [[ "$GLOBAL_TDD" == "TDD" || "$GLOBAL_TDD" == "tdd" ]]; then
-      TDD_MODE="TDD"
-      TDD_ENFORCEMENT="$GLOBAL_ENFORCEMENT"
-      TDD_SOURCE="global config"
-    fi
-  fi
-
-  # Step 2: Check active increment metadata (MEDIUM priority - overrides global)
-  ACTIVE_INCREMENT=""
-  for meta in "$SW_PROJECT_ROOT/.specweave/increments/"/*/metadata.json; do
-    [[ -f "$meta" ]] || continue
-    if jq -e '.status == "in-progress" or .status == "active"' "$meta" >/dev/null 2>&1; then
-      ACTIVE_INCREMENT="$meta"
-      break
-    fi
-  done
-
-  if [[ -n "$ACTIVE_INCREMENT" ]] && command -v jq >/dev/null 2>&1; then
-    INC_TDD=$(jq -r '.testMode // .tddMode // ""' "$ACTIVE_INCREMENT" 2>/dev/null)
-    INC_ENFORCEMENT=$(jq -r '.tddEnforcement // ""' "$ACTIVE_INCREMENT" 2>/dev/null)
-    if [[ "$INC_TDD" == "TDD" || "$INC_TDD" == "tdd" || "$INC_TDD" == "true" ]]; then
-      TDD_MODE="TDD"
-      [[ -n "$INC_ENFORCEMENT" ]] && TDD_ENFORCEMENT="$INC_ENFORCEMENT"
-      TDD_SOURCE="increment metadata"
-    elif [[ "$INC_TDD" == "test-after" || "$INC_TDD" == "false" ]]; then
-      # Increment explicitly disables TDD (overrides global)
-      TDD_MODE="off"
-      TDD_SOURCE="increment metadata (disabled)"
-    fi
-  fi
-
-  # Step 3: Check command flags in prompt (HIGHEST priority)
-  if echo "$PROMPT" | grep -qE '\-\-tdd|\-\-strict'; then
-    TDD_MODE="TDD"
-    TDD_ENFORCEMENT="strict"
-    TDD_SOURCE="command flag"
-  elif echo "$PROMPT" | grep -qE '\-\-no-tdd'; then
-    TDD_MODE="off"
-    TDD_SOURCE="command flag (disabled)"
-  fi
-
-  # Inject TDD context if enabled
-  if [[ "$TDD_MODE" == "TDD" ]]; then
-    ENFORCEMENT_DESC="warns but allows"
-    [[ "$TDD_ENFORCEMENT" == "strict" ]] && ENFORCEMENT_DESC="BLOCKS violations"
-
-    # v1.0.160: STRICT TDD adds mandatory blocking directive
-    if [[ "$TDD_ENFORCEMENT" == "strict" ]]; then
-      TDD_MSG="STRICT TDD ACTIVE (source: ${TDD_SOURCE}). RED->GREEN->REFACTOR enforced. No implementation before failing test. Use /sw:tdd-cycle."
-    fi
-
-    # v1.0.201: Include LSP instructions BEFORE TDD message
-    # This ensures LSP guidance is included in ALL early exits
-    if [[ -n "$LSP_EXPLICIT_REQUEST_MSG" ]]; then
-      if [[ -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-        AUTOLOAD_PLUGINS_MSG="${AUTOLOAD_PLUGINS_MSG}${LSP_EXPLICIT_REQUEST_MSG}"
-      else
-        AUTOLOAD_PLUGINS_MSG="$LSP_EXPLICIT_REQUEST_MSG"
-      fi
-    fi
-
-    # Append TDD message
-    if [[ -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-      AUTOLOAD_PLUGINS_MSG="${AUTOLOAD_PLUGINS_MSG}${TDD_MSG}"
-    else
-      AUTOLOAD_PLUGINS_MSG="$TDD_MSG"
-    fi
-
-    # Log TDD activation (use project root, never create dirs at $HOME)
-    if [[ -n "$SW_PROJECT_ROOT" ]]; then
-      TDD_LOG="$SW_PROJECT_ROOT/.specweave/logs/tdd-enforcement.log"
-      echo "[$(date -Iseconds)] TDD_MODE=$TDD_MODE | enforcement=$TDD_ENFORCEMENT | source=$TDD_SOURCE" >> "$TDD_LOG" 2>/dev/null
-    fi
-  fi
-fi
-
-# CRITICAL: Exit immediately for non-SpecWeave prompts
-# This covers 90%+ of prompts with <5ms overhead
-# v1.0.144: Still show plugin autoload message if plugins are being loaded
-# v1.0.155: AUTOLOAD_PLUGINS_MSG now includes new plugin warnings from helper function
-# v1.0.257: Expanded keywords to catch implementation prompts that bypass LLM detection
-if ! echo "$PROMPT" | grep -qiE "(specweave|/sw:|increment|add|create|implement|build|develop|test|component|feature|fix|refactor|write|style|setup|configure|migrate|deploy|scaffold)"; then
-  if [[ -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-    # Show plugin loading feedback even for non-SpecWeave prompts
-    output_approve_with_context "$AUTOLOAD_PLUGINS_MSG"
-  else
-    echo '{"decision":"approve"}'
-  fi
-  exit 0
-fi
-
-# ==============================================================================
-# EARLY EXIT FOR NON-SPECWEAVE PROJECTS (T-006 - v0.26.15)
-# ==============================================================================
-# Even if prompt contains SpecWeave keywords, exit if no .specweave directory
-# v1.0.144: Still show plugin autoload message for non-SpecWeave projects
-# Use SW_PROJECT_ROOT to avoid creating dirs relative to CWD
-if [[ -n "$SW_PROJECT_ROOT" ]]; then
-  SPECWEAVE_DIR="$SW_PROJECT_ROOT/.specweave"
-else
-  SPECWEAVE_DIR=".specweave"
-fi
-if [[ ! -d "$SPECWEAVE_DIR" ]]; then
-  if [[ -n "$AUTOLOAD_PLUGINS_MSG" ]]; then
-    # Show plugin loading feedback even for non-SpecWeave projects
-    output_approve_with_context "$AUTOLOAD_PLUGINS_MSG"
-  else
-    echo '{"decision":"approve"}'
-  fi
-  exit 0
-fi
-
-# ==============================================================================
-# INSTANT SCRIPT EXECUTION: Status commands bypass LLM entirely (v0.33.0)
-# ==============================================================================
-# These commands need NO LLM reasoning - execute scripts directly for <1s response
-# Pattern: Detect command → Execute script → Return output via "block" → Exit
-
-PLUGIN_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-SCRIPTS_DIR="$PLUGIN_ROOT/scripts"
-
-# Helper: Escape output for JSON (handles newlines, quotes, backslashes)
-# Uses jq for proper JSON string escaping (required dependency for instant commands)
-escape_json() {
-  local input="$1"
-  # jq -Rs properly escapes all special characters including newlines
-  # We strip the surrounding quotes since we add them in the printf
-  printf '%s' "$input" | jq -Rs '.' | sed 's/^"//; s/"$//'
-}
-
-# NOTE: output_approve_with_context() is defined earlier in the file (line ~105)
-# It uses hookSpecificOutput.additionalContext (NOT systemMessage!)
-
-# Helper: Check if running in VSCode extension
-# VSCode sets CLAUDE_CODE_ENTRYPOINT=claude-vscode
-# Returns 0 (true) if VSCode, 1 (false) if CLI
-is_vscode() {
-  [[ -n "${CLAUDE_CODE_ENTRYPOINT}" ]] && [[ "${CLAUDE_CODE_ENTRYPOINT}" == "claude-vscode" ]]
-}
-
-# Helper: Extract command line and args from multi-line prompt (v1.0.105+)
-# When prompts contain IDE metadata (e.g., <ide_opened_file>...</ide_opened_file>)
-# the command may be on a subsequent line. This function:
-# 1. Finds the line containing the command
-# 2. Extracts args from that specific line
-# Usage: extract_command_args "PROMPT" "command_pattern" (e.g., "/sw:progress")
-# Returns args on stdout, or empty string if no args
-extract_command_args() {
-  local prompt="$1"
-  local cmd_pattern="$2"
-
-  # Find the line containing the command and extract args from it
-  # The grep -oE gets just the matching line, sed removes the command prefix
-  local cmd_line
-  cmd_line=$(echo "$prompt" | grep -E "^${cmd_pattern}($| )" | head -1)
-
-  if [[ -n "$cmd_line" ]]; then
-    # Remove the command pattern from the line to get args
-    echo "$cmd_line" | sed "s|^${cmd_pattern}[[:space:]]*||"
-  fi
-}
-
-# /sw:jobs → Execute read-jobs.sh (pure bash, ~2ms)
-if echo "$PROMPT" | grep -qE "^/sw:jobs($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:jobs")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-jobs.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-jobs.sh" "$ARGS" 2>&1)
-  elif [[ -f "$SCRIPTS_DIR/jobs.js" ]] && command -v node >/dev/null 2>&1; then
-    OUTPUT=$(cd "$(pwd)" && node "$SCRIPTS_DIR/jobs.js" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No jobs script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:progress → Execute read-progress.sh (pure bash, ~30ms)
-if echo "$PROMPT" | grep -qE "^/sw:progress($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:progress")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-progress.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-progress.sh" "$ARGS" 2>&1)
-  elif [[ -f "$SCRIPTS_DIR/progress.js" ]] && command -v node >/dev/null 2>&1; then
-    OUTPUT=$(cd "$(pwd)" && node "$SCRIPTS_DIR/progress.js" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No progress script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:grill → Load increment context for code review (pure bash, ~50ms)
-# Unlike /sw:progress which displays data, this injects CONTEXT + INSTRUCTIONS
-# so the LLM performs a structured code review with full increment awareness.
-if echo "$PROMPT" | grep -qE "^/sw:grill($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:grill")
-
-  if [[ -f "$SCRIPTS_DIR/read-grill-context.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-grill-context.sh" $ARGS 2>&1)
-  else
-    OUTPUT="GRILL MODE ACTIVATED — Run a thorough code review of the active increment. Check correctness, security, performance, and maintainability. Categorize issues as BLOCKER, CRITICAL, MAJOR, MINOR, or SUGGESTION. End with VERDICT: PASS or FAIL."
-  fi
-
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:status → Execute read-status.sh (pure bash, ~150ms)
-if echo "$PROMPT" | grep -qE "^/sw:status($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:status")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-status.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-status.sh" "$ARGS" 2>&1)
-  elif [[ -f "$SCRIPTS_DIR/status.js" ]] && command -v node >/dev/null 2>&1; then
-    OUTPUT=$(cd "$(pwd)" && node "$SCRIPTS_DIR/status.js" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No status script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:workflow → Execute read-workflow.sh (pure bash, ~100ms)
-if echo "$PROMPT" | grep -qE "^/sw:workflow($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:workflow")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-workflow.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-workflow.sh" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No workflow script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:costs → Execute read-costs.sh (pure bash, ~50ms)
-if echo "$PROMPT" | grep -qE "^/sw:costs($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:costs")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-costs.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-costs.sh" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No costs script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# /sw:analytics → Execute read-analytics.sh (pure bash, ~50ms)
-if echo "$PROMPT" | grep -qE "^/sw:analytics($| )"; then
-  ARGS=$(extract_command_args "$PROMPT" "/sw:analytics")
-
-  # Execute command and get output
-  if [[ -f "$SCRIPTS_DIR/read-analytics.sh" ]]; then
-    OUTPUT=$(cd "$(pwd)" && bash "$SCRIPTS_DIR/read-analytics.sh" "$ARGS" 2>&1)
-  else
-    OUTPUT="❌ No analytics script available"
-  fi
-
-  # Unified response for both CLI and VSCode (v1.0.166)
-  # Uses additionalContext (NOT systemMessage) to inject output into Claude's context
-  output_approve_with_context "$OUTPUT"
-  exit 0
-fi
-
-# ==============================================================================
-# TASK COUNT GUARD: Block /sw:do for oversized increments (v0.32.2+)
-# ==============================================================================
-# >8 tasks = context explosion = CRASH (per CLAUDE.md)
-MAX_TASKS=8
-
-if echo "$PROMPT" | grep -qE "^/sw:do($| )"; then
-  # Extract increment ID from prompt
-  DO_INCREMENT_ID=$(echo "$PROMPT" | grep -oE "[0-9]{4}[a-zA-Z0-9-]*" | head -1)
-
-  # If no ID provided, find active increment
-  if [[ -z "$DO_INCREMENT_ID" ]]; then
-    for meta in "$SPECWEAVE_DIR/increments"/*/metadata.json; do
-      [[ -f "$meta" ]] || continue
-      if command -v jq >/dev/null 2>&1; then
-        status=$(jq -r '.status // "unknown"' "$meta" 2>/dev/null)
-      else
-        status=$(sed -n 's/.*"status"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' "$meta" 2>/dev/null || echo "unknown")
-      fi
-      if [[ "$status" == "active" || "$status" == "planning" || "$status" == "backlog" || "$status" == "ready_for_review" ]]; then
-        DO_INCREMENT_ID=$(basename "$(dirname "$meta")")
-        break
-      fi
-    done
-  fi
-
-  if [[ -n "$DO_INCREMENT_ID" ]]; then
-    TASKS_FILE="$SPECWEAVE_DIR/increments/$DO_INCREMENT_ID/tasks.md"
-    if [[ -f "$TASKS_FILE" ]]; then
-      TASK_COUNT=$(grep -c "^### T-" "$TASKS_FILE" 2>/dev/null || echo "0")
-      if [[ "$TASK_COUNT" -gt "$MAX_TASKS" ]]; then
-        printf '{"decision":"block","reason":"❌ TASK COUNT EXCEEDS LIMIT\\n\\nIncrement %s has %s tasks (maximum: %s)\\n\\n>8 tasks = context explosion = CRASH\\n\\n💡 REQUIRED: Split into smaller increments:\\n\\n  Pattern: %s/ → Split into:\\n    • %s-part1/ (T-001 to T-004)\\n    • Next increment (T-005 to T-008)\\n    • Next increment (T-009+)\\n\\n⚠️ DO NOT PROCEED until tasks.md has ≤8 tasks!"}\n' "$DO_INCREMENT_ID" "$TASK_COUNT" "$MAX_TASKS" "$DO_INCREMENT_ID" "$DO_INCREMENT_ID"
-        exit 0
-      fi
-    fi
-  fi
-fi
-
-# ==============================================================================
-# CACHED ACTIVE INCREMENT DETECTION (ONCE - reused throughout!)
-# ==============================================================================
-ACTIVE_INCREMENT=""
-ACTIVE_COUNT=0
-ACTIVE_LIST=""
-
-if [[ -d "$SPECWEAVE_DIR/increments" ]]; then
-  # Single find + jq pass to get ALL active increment info
-  while IFS= read -r metadata_file; do
-    [[ -z "$metadata_file" ]] && continue
-
-    # Use jq (fast) to extract status and id
-    if command -v jq >/dev/null 2>&1; then
-      read -r status inc_type < <(jq -r '"\(.status // "unknown") \(.type // "feature")"' "$metadata_file" 2>/dev/null || echo "unknown feature")
-    else
-      # Fallback: grep (no node!)
-      status=$(sed -n 's/.*"status"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' "$metadata_file" 2>/dev/null || echo "unknown")
-      inc_type=$(sed -n 's/.*"type"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' "$metadata_file" 2>/dev/null || echo "feature")
-    fi
-
-    if [[ "$status" == "active" || "$status" == "planning" || "$status" == "ready_for_review" ]]; then
-      inc_id=$(basename "$(dirname "$metadata_file")")
-      ACTIVE_COUNT=$((ACTIVE_COUNT + 1))
-      ACTIVE_LIST="${ACTIVE_LIST}  - $inc_id [$inc_type]\n"
-      [[ -z "$ACTIVE_INCREMENT" ]] && ACTIVE_INCREMENT="$inc_id"
-    fi
-  done < <(find "$SPECWEAVE_DIR/increments" -maxdepth 2 -name "metadata.json" -not -path "*/_archive/*" 2>/dev/null)
-fi
-
-# ==============================================================================
-# WIP WARNING BUILDER (reusable across all increment suggestion paths)
-# ==============================================================================
-# Builds a soft warning string when active increments >= configured limit.
-# Never blocks — just informs. Prepended to increment suggestion messages.
-WIP_WARNING=""
-if [[ "$ACTIVE_COUNT" -gt 0 ]]; then
-  _WIP_CONFIG="${SPECWEAVE_DIR}/config.json"
-  _WIP_LIMIT=3
-  if [[ -f "$_WIP_CONFIG" ]] && command -v jq >/dev/null 2>&1; then
-    _WIP_LIMIT=$(jq -r '.limits.maxActiveIncrements // 3' "$_WIP_CONFIG" 2>/dev/null || echo "3")
-  fi
-  [[ ! "$_WIP_LIMIT" =~ ^[0-9]+$ ]] && _WIP_LIMIT=3
-
-  if [[ "$ACTIVE_COUNT" -ge "$_WIP_LIMIT" ]]; then
-    WIP_WARNING="⚠️ **WIP Notice** (${ACTIVE_COUNT}/${_WIP_LIMIT} active)\\n\\nActive increments:\\n${ACTIVE_LIST}\\nConsider completing existing work first (\`/sw:done <id>\`) or pausing (\`specweave pause <id>\`).\\n\\n---\\n\\n"
-  fi
-fi
-
-# ==============================================================================
-# ARCHIVE SUGGESTION (v1.0.257 - Auto-archive when too many increments)
-# ==============================================================================
-# When total numbered increment directories exceed threshold, suggest archiving.
-# In auto mode: archive silently. Interactive: inject suggestion for LLM.
-# Rate-limited: once per day via marker file.
-ARCHIVE_SUGGESTION_MSG=""
-if [[ -d "$SPECWEAVE_DIR/increments" ]]; then
-  TOTAL_INCREMENT_DIRS=$(ls -d "$SPECWEAVE_DIR/increments"/[0-9]* 2>/dev/null | wc -l | tr -d ' ')
-
-  # Read threshold from config (default: 10)
-  _ARCHIVE_THRESHOLD=10
-  if [[ -f "$CONFIG_PATH" ]] && command -v jq >/dev/null 2>&1; then
-    _ARCHIVE_THRESHOLD=$(jq -r '.archiving.autoArchiveThreshold // 10' "$CONFIG_PATH" 2>/dev/null || echo "10")
-  fi
-  [[ ! "$_ARCHIVE_THRESHOLD" =~ ^[0-9]+$ ]] && _ARCHIVE_THRESHOLD=10
-
-  if [[ "$TOTAL_INCREMENT_DIRS" -ge "$_ARCHIVE_THRESHOLD" ]]; then
-    # Rate limit: once per day via marker file
-    ARCHIVE_MARKER="$SPECWEAVE_DIR/state/archive-suggestion.marker"
-    SHOULD_SUGGEST=true
-    if [[ -f "$ARCHIVE_MARKER" ]]; then
-      MARKER_DATE=$(cat "$ARCHIVE_MARKER" 2>/dev/null | head -1)
-      TODAY=$(date +%Y-%m-%d)
-      [[ "$MARKER_DATE" == "$TODAY" ]] && SHOULD_SUGGEST=false
-    fi
-
-    if [[ "$SHOULD_SUGGEST" == "true" ]]; then
-      # Check if in auto mode
-      AUTO_STATE_FILE="$SPECWEAVE_DIR/state/auto.json"
-      IN_AUTO_MODE=false
-      if [[ -f "$AUTO_STATE_FILE" ]] && command -v jq >/dev/null 2>&1; then
-        AUTO_STATUS=$(jq -r '.status // "idle"' "$AUTO_STATE_FILE" 2>/dev/null)
-        [[ "$AUTO_STATUS" == "running" ]] && IN_AUTO_MODE=true
-      fi
-
-      if [[ "$IN_AUTO_MODE" == "true" ]]; then
-        # Auto mode: archive silently in background
-        if command -v specweave >/dev/null 2>&1; then
-          specweave archive --keep-last 10 2>/dev/null &
-        fi
-      else
-        # Interactive: suggest to LLM
-        ARCHIVE_SUGGESTION_MSG="⚠️ **Archive suggested**: ${TOTAL_INCREMENT_DIRS} increments in workspace (threshold: ${_ARCHIVE_THRESHOLD}). Run: \`specweave archive --keep-last 10\`\n\n"
-      fi
-
-      # Write today's date as marker
-      mkdir -p "$(dirname "$ARCHIVE_MARKER")" 2>/dev/null
-      echo "$(date +%Y-%m-%d)" > "$ARCHIVE_MARKER" 2>/dev/null || true
-    fi
-  fi
-fi
-
-# ==============================================================================
-# SMART INTERVIEW GATE (v1.0.243 - LLM-Driven Prompt Assessment)
-# ==============================================================================
-# When Deep Interview Mode is enabled AND no active increment exists,
-# inject instructions for the LLM to assess prompt completeness.
-# The LLM decides: ask targeted questions OR proceed to increment creation.
-# Fires on EVERY prompt until an increment is created.
-# See ADR-0243 for architecture decision.
-
-SMART_INTERVIEW_GATE_MSG=""
-if [[ "$DEEP_INTERVIEW_ENABLED" == "true" ]] && [[ -z "$ACTIVE_INCREMENT" ]]; then
-  # Also check active-increment.json state file as secondary source
-  HAVE_ACTIVE_STATE=false
-  STATE_FILE="$SPECWEAVE_DIR/state/active-increment.json"
-  if [[ -f "$STATE_FILE" ]] && command -v jq >/dev/null 2>&1; then
-    ACTIVE_IDS=$(jq -r '.ids // [] | length' "$STATE_FILE" 2>/dev/null || echo "0")
-    [[ "$ACTIVE_IDS" -gt 0 ]] && HAVE_ACTIVE_STATE=true
-  fi
-
-  if [[ "$HAVE_ACTIVE_STATE" != "true" ]]; then
-    SMART_INTERVIEW_GATE_MSG="No active increment. Assess prompt completeness for complexity — if gaps, ask targeted questions (count depends on complexity). If sufficient, call sw:increment."
-  fi
-fi
-
-# ==============================================================================
-# DISCIPLINE VALIDATION: Warn about WIP limits (configurable, not hard block!)
-# ==============================================================================
-
-# ==============================================================================
-# PROJECT CONTEXT + WIP LIMITS FOR /sw:increment (v0.34.0)
-# ==============================================================================
-# CRITICAL: Inject project/board context BEFORE Claude generates spec.md
-# This ensures Claude knows available projects and uses correct IDs
-# ALSO: Check WIP limits in same block to avoid duplicate command detection
-
-if echo "$PROMPT" | grep -qE "^/sw:increment"; then
-  # Get project context (uses specweave CLI if available)
-  PROJECT_CONTEXT=""
-
-  if command -v specweave >/dev/null 2>&1; then
-    # Use CLI for accurate project/board detection
-    CONTEXT_JSON=$(specweave context projects 2>/dev/null || echo '{}')
-
-    # Validate JSON before parsing (defensive coding)
-    if [[ -n "$CONTEXT_JSON" ]] && [[ "$CONTEXT_JSON" != "{}" ]]; then
-      if command -v jq >/dev/null 2>&1; then
-        # Verify JSON is parseable before extracting fields
-        if ! echo "$CONTEXT_JSON" | jq empty 2>/dev/null; then
-          CONTEXT_JSON='{}'  # Invalid JSON - reset to empty
-        fi
-      fi
-    fi
-
-    if [[ -n "$CONTEXT_JSON" ]] && [[ "$CONTEXT_JSON" != "{}" ]]; then
-      # Parse JSON with jq
-      if command -v jq >/dev/null 2>&1; then
-        LEVEL=$(echo "$CONTEXT_JSON" | jq -r '.level // 1')
-        PROJECTS=$(echo "$CONTEXT_JSON" | jq -r '.projects | map(.id) | join(", ")' 2>/dev/null || echo "")
-
-        if [[ "$LEVEL" == "2" ]]; then
-          # 2-level structure: include boards
-          BOARDS_JSON=$(echo "$CONTEXT_JSON" | jq -r '.boardsByProject // {}' 2>/dev/null)
-          if [[ -n "$BOARDS_JSON" ]] && [[ "$BOARDS_JSON" != "{}" ]]; then
-            PROJECT_CONTEXT="\\n\\n📦 PROJECT CONTEXT (2-LEVEL STRUCTURE)\\n\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}⚠️ MANDATORY: Each User Story MUST have both:\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}  - **Project**: <project_id>\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}  - **Board**: <board_id>\\n\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}Available projects: ${PROJECTS}\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}Boards by project:\\n"
-
-            # Format boards
-            for proj in $(echo "$CONTEXT_JSON" | jq -r '.projects[].id' 2>/dev/null); do
-              PROJ_BOARDS=$(echo "$CONTEXT_JSON" | jq -r ".boardsByProject[\"$proj\"] | map(.id) | join(\", \")" 2>/dev/null || echo "")
-              [[ -n "$PROJ_BOARDS" ]] && PROJECT_CONTEXT="${PROJECT_CONTEXT}  - ${proj}: ${PROJ_BOARDS}\\n"
-            done
-
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}\\n❌ FORBIDDEN: Comma-separated values (e.g., **Project**: fe, be)\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}✅ REQUIRED: One project + one board per User Story"
-          fi
-        elif [[ -n "$PROJECTS" ]]; then
-          # 1-level structure: projects only
-          PROJECT_COUNT=$(echo "$CONTEXT_JSON" | jq '.projects // [] | length' 2>/dev/null || echo "0")
-
-          if [[ "$PROJECT_COUNT" -gt 1 ]]; then
-            PROJECT_CONTEXT="\\n\\n📦 PROJECT CONTEXT (MULTI-PROJECT)\\n\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}⚠️ MANDATORY: Each User Story MUST have:\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}  - **Project**: <project_id>\\n\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}Available projects: ${PROJECTS}\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}\\n❌ FORBIDDEN: Comma-separated values\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}✅ REQUIRED: One project per User Story"
-          elif [[ "$PROJECT_COUNT" -eq 1 ]]; then
-            # Single project: auto-select
-            SINGLE_PROJECT=$(echo "$CONTEXT_JSON" | jq -r '.projects[0].id' 2>/dev/null)
-            PROJECT_CONTEXT="\\n\\n📦 PROJECT CONTEXT\\n"
-            PROJECT_CONTEXT="${PROJECT_CONTEXT}Single project detected: ${SINGLE_PROJECT} (auto-selected)"
-          fi
-        fi
-      fi
-    fi
-  else
-    # Fallback: Check for multi-project folders
-    if [[ -d "$SPECWEAVE_DIR/docs/internal/specs" ]]; then
-      PROJ_COUNT=$(find "$SPECWEAVE_DIR/docs/internal/specs" -maxdepth 1 -type d | wc -l)
-      if [[ "$PROJ_COUNT" -gt 2 ]]; then
-        PROJ_LIST=$(ls -1 "$SPECWEAVE_DIR/docs/internal/specs" 2>/dev/null | grep -v "_" | tr '\n' ', ' | sed 's/,$//')
-        PROJECT_CONTEXT="\\n\\n📦 PROJECT CONTEXT (MULTI-PROJECT)\\n"
-        PROJECT_CONTEXT="${PROJECT_CONTEXT}⚠️ MANDATORY: Each User Story MUST have **Project**: field\\n"
-        PROJECT_CONTEXT="${PROJECT_CONTEXT}Available folders: ${PROJ_LIST}"
-      fi
-    fi
-  fi
-
-  # WIP + PROJECT CONTEXT: Combine WIP_WARNING (built earlier) with project context
-  # Never blocks — always approve with informational context
-  COMBINED_CONTEXT="${WIP_WARNING}${PROJECT_CONTEXT}"
-  if [[ -n "$COMBINED_CONTEXT" ]]; then
-    output_approve_with_context "$COMBINED_CONTEXT"
-    exit 0
-  fi
-fi
-
-# ==============================================================================
-# PRE-FLIGHT SYNC CHECK (LIGHTWEIGHT - uses cached ACTIVE_INCREMENT)
-# ==============================================================================
-
-# Detect increment operations that need fresh data
-if echo "$PROMPT" | grep -qE "/(specweave:)?(done|validate|progress|do)"; then
-  # Extract increment ID from prompt OR use cached active
-  INCREMENT_ID=$(echo "$PROMPT" | grep -oE "[0-9]{4}[a-z0-9-]*" | head -1)
-  [[ -z "$INCREMENT_ID" ]] && INCREMENT_ID="$ACTIVE_INCREMENT"
-
-  # If we have an increment ID, check freshness (pure bash - no node!)
-  if [[ -n "$INCREMENT_ID" ]]; then
-    INCREMENT_SPEC="$SPECWEAVE_DIR/increments/$INCREMENT_ID/spec.md"
-    LIVING_DOCS_SPEC="$SPECWEAVE_DIR/docs/internal/specs/spec-$INCREMENT_ID.md"
-
-    if [[ -f "$INCREMENT_SPEC" ]]; then
-      # Use find -newer for mtime comparison (single syscall!)
-      if [[ ! -f "$LIVING_DOCS_SPEC" ]] || [[ -n $(find "$INCREMENT_SPEC" -newer "$LIVING_DOCS_SPEC" 2>/dev/null) ]]; then
-        # Sync needed - run async (non-blocking!)
-        PLUGIN_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-        SYNC_SCRIPT="$PLUGIN_ROOT/lib/hooks/sync-living-docs.js"
-        [[ -f "$SYNC_SCRIPT" ]] && node "$SYNC_SCRIPT" "$INCREMENT_ID" >/dev/null 2>&1 &
-      fi
-    fi
-  fi
-fi
-
-# ==============================================================================
-# SPEC SYNC CHECK (LIGHTWEIGHT - only when really needed)
-# ==============================================================================
-# Skip SpecSyncManager for most prompts - it's HEAVY!
-# Only check on explicit sync-related commands
-
-if [[ -n "$ACTIVE_INCREMENT" ]] && echo "$PROMPT" | grep -qE "/(specweave:)?(sync|done)"; then
-  # Simple mtime check: spec.md vs plan.md (pure bash!)
-  SPEC_FILE="$SPECWEAVE_DIR/increments/$ACTIVE_INCREMENT/spec.md"
-  PLAN_FILE="$SPECWEAVE_DIR/increments/$ACTIVE_INCREMENT/plan.md"
-
-  if [[ -f "$SPEC_FILE" ]] && [[ -f "$PLAN_FILE" ]]; then
-    # Check if spec is newer than plan (indicates spec changes need sync)
-    if [[ -n $(find "$SPEC_FILE" -newer "$PLAN_FILE" 2>/dev/null) ]]; then
-      output_approve_with_context "⚠️ Spec changes detected in ${ACTIVE_INCREMENT}\n\nspec.md has been modified after plan.md.\nConsider running /sw:sync-docs to update living documentation."
-      exit 0
-    fi
-  fi
-fi
-
-# ==============================================================================
-# CONTEXT INJECTION (uses cached ACTIVE_INCREMENT - no more find loops!)
-# ==============================================================================
-
-CONTEXT=""
-
-if [[ -n "$ACTIVE_INCREMENT" ]]; then
-  # Read from status-line.json cache (single source of truth)
-  CACHE_FILE="$SPECWEAVE_DIR/state/status-line.json"
-
-  if [[ -f "$CACHE_FILE" ]]; then
-    # Single jq call for all values (or pure bash fallback)
-    if command -v jq >/dev/null 2>&1; then
-      read -r TOTAL_TASKS COMPLETED_TASKS TOTAL_ACS COMPLETED_ACS < <(
-        jq -r '[.current.total // 0, .current.completed // 0, .current.acsTotal // 0, .current.acsCompleted // 0] | @tsv' "$CACHE_FILE" 2>/dev/null || echo "0 0 0 0"
-      )
-    else
-      # Pure grep fallback (no node!)
-      TOTAL_TASKS=$(sed -n 's/.*"total"[[:space:]]*:[[:space:]]*\([0-9][0-9]*\).*/\1/p' "$CACHE_FILE" 2>/dev/null | head -1 || echo "0")
-      COMPLETED_TASKS=$(sed -n 's/.*"completed"[[:space:]]*:[[:space:]]*\([0-9][0-9]*\).*/\1/p' "$CACHE_FILE" 2>/dev/null | head -1 || echo "0")
-      TOTAL_ACS=$(sed -n 's/.*"acsTotal"[[:space:]]*:[[:space:]]*\([0-9][0-9]*\).*/\1/p' "$CACHE_FILE" 2>/dev/null || echo "0")
-      COMPLETED_ACS=$(sed -n 's/.*"acsCompleted"[[:space:]]*:[[:space:]]*\([0-9][0-9]*\).*/\1/p' "$CACHE_FILE" 2>/dev/null || echo "0")
-    fi
-
-    # Ensure valid numbers
-    TOTAL_TASKS=${TOTAL_TASKS:-0}
-    COMPLETED_TASKS=${COMPLETED_TASKS:-0}
-    TOTAL_ACS=${TOTAL_ACS:-0}
-    COMPLETED_ACS=${COMPLETED_ACS:-0}
-
-    if [[ "$TOTAL_TASKS" -gt 0 ]] 2>/dev/null; then
-      PERCENTAGE=$(( COMPLETED_TASKS * 100 / TOTAL_TASKS ))
-
-      if [[ "$TOTAL_ACS" -gt 0 ]] 2>/dev/null; then
-        AC_PERCENTAGE=$(( COMPLETED_ACS * 100 / TOTAL_ACS ))
-        CONTEXT="✓ Active: $ACTIVE_INCREMENT ($COMPLETED_TASKS/$TOTAL_TASKS tasks, $PERCENTAGE% | $COMPLETED_ACS/$TOTAL_ACS ACs, $AC_PERCENTAGE%)"
-      else
-        CONTEXT="✓ Active: $ACTIVE_INCREMENT ($COMPLETED_TASKS/$TOTAL_TASKS tasks, $PERCENTAGE%)"
-      fi
-    else
-      CONTEXT="✓ Active: $ACTIVE_INCREMENT"
-    fi
-  else
-    CONTEXT="✓ Active: $ACTIVE_INCREMENT"
-  fi
-fi
-
-# ==============================================================================
-# COMMAND SUGGESTIONS: Guide users to structured workflow
-# ==============================================================================
-
-# Command suggestions removed (v1.0.257) — already in CLAUDE.md, reduces per-turn context
-
-# ==============================================================================
-# STATUS LINE REFRESH (v0.26.13 - CONDITIONAL + ASYNC)
-# ==============================================================================
-# Only refresh when we have an active increment (skip for most prompts)
-# Runs in background to avoid blocking user prompt
-
-if [[ -n "$ACTIVE_INCREMENT" ]] && [[ -d "$SPECWEAVE_DIR" ]]; then
-  HOOK_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-  # Run async (non-blocking!) - update-status-line.sh has its own TTL/mtime guards
-  bash "$HOOK_DIR/lib/update-status-line.sh" 2>/dev/null &
-fi
-
-# ==============================================================================
-# ANALYTICS TRACKING: Track /sw:* command invocations (v0.35.0+)
-# ==============================================================================
-# Non-blocking, runs in background. Tracks command usage for /sw:analytics.
-
-if echo "$PROMPT" | grep -qE "^/sw:[a-z]"; then
-  COMMAND_NAME=$(echo "$PROMPT" | grep -oE "^/sw:[a-z0-9:-]+" | head -1)
-  if [[ -n "$COMMAND_NAME" ]] && [[ -f "$SCRIPTS_DIR/track-analytics.sh" ]]; then
-    bash "$SCRIPTS_DIR/track-analytics.sh" command "$COMMAND_NAME" --plugin specweave --increment "$ACTIVE_INCREMENT" 2>/dev/null &
-  fi
-fi
-
-# ==============================================================================
-# OUTPUT: Priority-based context assembly with budget (v1.0.260)
-# ==============================================================================
-# Assembles final message by adding context items in priority order,
-# stopping when the budget is exhausted. This prevents blind concatenation
-# that wastes budget on low-priority items while truncating critical ones.
-#
-# Priority tiers:
-#   P1 (critical): Plugin status (RESTART/Using), active increment status
-#   P2 (important): LSP explicit request, WIP/interview gate
-#   P3 (informational): LSP setup/install suggestions, archive suggestion
-
-# ==============================================================================
-# CONTEXT BUDGET RESOLUTION (v1.0.262)
-# ==============================================================================
-# 1. Read base level from config.json (contextBudget.level)
-# 2. If autoAdapt=true, check context-pressure.json and step down
-# 3. Map level to char budget
-# 4. Environment override: SPECWEAVE_CONTEXT_BUDGET
-
-BUDGET_LEVEL="full"
-AUTO_ADAPT=true
-if [[ -f "$CONFIG_PATH" ]] && command -v jq >/dev/null 2>&1; then
-  BUDGET_LEVEL=$(jq -r '.contextBudget.level // "full"' "$CONFIG_PATH" 2>/dev/null || echo "full")
-  AUTO_ADAPT_VAL=$(jq -r '.contextBudget.autoAdapt // true' "$CONFIG_PATH" 2>/dev/null || echo "true")
-  [[ "$AUTO_ADAPT_VAL" == "false" ]] && AUTO_ADAPT=false
-fi
-
-# Environment override (for quick testing without config changes)
-[[ -n "${SPECWEAVE_CONTEXT_BUDGET:-}" ]] && BUDGET_LEVEL="$SPECWEAVE_CONTEXT_BUDGET"
-
-# Auto-adapt: check pressure state from PreCompact hook
-if [[ "$AUTO_ADAPT" == "true" ]] && [[ -n "$SW_PROJECT_ROOT" ]]; then
-  PRESSURE_FILE="$SW_PROJECT_ROOT/.specweave/state/context-pressure.json"
-  if [[ -f "$PRESSURE_FILE" ]] && command -v jq >/dev/null 2>&1; then
-    PRESSURE_LEVEL=$(jq -r '.level // "normal"' "$PRESSURE_FILE" 2>/dev/null || echo "normal")
-    case "$PRESSURE_LEVEL" in
-      elevated)
-        # Step down one level
-        case "$BUDGET_LEVEL" in
-          full) BUDGET_LEVEL="compact" ;;
-          compact) BUDGET_LEVEL="minimal" ;;
-        esac
-        ;;
-      critical)
-        # Jump to minimal regardless of base
-        [[ "$BUDGET_LEVEL" != "off" ]] && BUDGET_LEVEL="minimal"
-        ;;
-      emergency)
-        # Nuclear option: strip ALL context (v1.0.268 - 3+ compactions)
-        BUDGET_LEVEL="off"
-        ;;
-    esac
-  fi
-fi
-
-# Map level to char budget
-case "$BUDGET_LEVEL" in
-  full)    CONTEXT_BUDGET=2500 ;;
-  compact) CONTEXT_BUDGET=1000 ;;
-  minimal) CONTEXT_BUDGET=300  ;;
-  off)     CONTEXT_BUDGET=0    ;;
-  *)       CONTEXT_BUDGET=2500 ;;
-esac
-
-# If budget is 0, send remediation message if pressure-triggered, then exit
-if [[ "$CONTEXT_BUDGET" -eq 0 ]]; then
-  if [[ "${PRESSURE_LEVEL:-}" == "emergency" || "${PRESSURE_LEVEL:-}" == "critical" ]]; then
-    output_approve_with_context "Context budget auto-reduced due to prompt pressure. Consider starting a fresh session."
-  else
-    echo '{"decision":"approve"}'
-  fi
-  exit 0
-fi
-
-# Helper: Append message to FINAL_MESSAGE if it fits within budget
-# Args: $1=message to append
-# Returns: 0 if appended, 1 if skipped (budget exceeded)
-_budget_append() {
-  local msg="$1"
-  [[ -z "$msg" ]] && return 0
-  local new_len=$(( ${#FINAL_MESSAGE} + ${#msg} ))
-  if [[ $new_len -le $CONTEXT_BUDGET ]]; then
-    FINAL_MESSAGE="${FINAL_MESSAGE}${msg}"
-    return 0
-  fi
-  return 1
-}
-
-FINAL_MESSAGE=""
-
-# P1: Critical — plugin status and active increment
-_budget_append "$AUTOLOAD_PLUGINS_MSG"
-_budget_append "$CONTEXT"
-
-# P2: Important — LSP explicit request, interview gate
-_budget_append "$LSP_EXPLICIT_REQUEST_MSG"
-if [[ -n "$SMART_INTERVIEW_GATE_MSG" ]]; then
-  _budget_append "\\n${SMART_INTERVIEW_GATE_MSG}"
-fi
-
-# P3: Informational — LSP setup, archive, environment
-_budget_append "$LSP_ENV_SETUP_MSG"
-_budget_append "$LSP_INSTALL_MSG"
-_budget_append "$LSP_SETUP_SUGGESTION_MSG"
-_budget_append "$ARCHIVE_SUGGESTION_MSG"
-
-# ==============================================================================
-# TURN DEDUPLICATION: Skip identical context to save tokens (v1.0.262)
-# ==============================================================================
-# If this turn's context is identical to last turn's, don't re-inject it.
-# Claude already has it in history. Saves ~2500 chars per duplicate turn.
-if [[ -n "$FINAL_MESSAGE" ]] && [[ -n "$SW_PROJECT_ROOT" ]]; then
-  DEDUP_HASH_FILE="$SW_PROJECT_ROOT/.specweave/state/.context-hash"
-  CURRENT_HASH=""
-  # SMART_INTERVIEW_GATE_MSG is excluded from the dedup hash so the gate
-  # fires on every turn even when the rest of the context is identical.
-  HASH_INPUT="$FINAL_MESSAGE"
-  if [[ -n "$SMART_INTERVIEW_GATE_MSG" ]]; then
-    HASH_INPUT="${HASH_INPUT//$SMART_INTERVIEW_GATE_MSG/}"
-  fi
-  if command -v md5sum >/dev/null 2>&1; then
-    CURRENT_HASH=$(printf '%s' "$HASH_INPUT" | md5sum | cut -d' ' -f1)
-  elif command -v md5 >/dev/null 2>&1; then
-    CURRENT_HASH=$(printf '%s' "$HASH_INPUT" | md5)
-  fi
-
-  if [[ -n "$CURRENT_HASH" ]]; then
-    if [[ -f "$DEDUP_HASH_FILE" ]]; then
-      PREV_HASH=$(cat "$DEDUP_HASH_FILE" 2>/dev/null)
-      if [[ "$CURRENT_HASH" == "$PREV_HASH" ]]; then
-        # Identical to last turn — skip injection
-        echo '{"decision":"approve"}'
-        exit 0
-      fi
-    fi
-    # Save hash for next turn
-    echo "$CURRENT_HASH" > "$DEDUP_HASH_FILE" 2>/dev/null
-  fi
-fi
-
-if [[ -n "$FINAL_MESSAGE" ]]; then
-  output_approve_with_context "$FINAL_MESSAGE"
-else
-  echo '{"decision":"approve"}'
-fi
-
-exit 0