diff --git a/.agents/scripts/commands/cross-review.md b/.agents/scripts/commands/cross-review.md
index 0e4ee21233..c6f0342f9f 100644
--- a/.agents/scripts/commands/cross-review.md
+++ b/.agents/scripts/commands/cross-review.md
@@ -1,30 +1,21 @@
 ---
-description: Dispatch a prompt to multiple AI models, diff results, and optionally score via a judge model
+description: Dispatch the same prompt to multiple AI models, diff results, and optionally auto-score via a judge model
 agent: Build+
 mode: subagent
 ---
 
-Run a multi-model adversarial review: dispatch the same prompt to N models in parallel, collect outputs, diff results, and optionally score via a judge model (Ouroboros-style pipeline).
+Dispatch a prompt to multiple AI models in parallel, collect and diff their responses, and optionally score them via a judge model.
 
 Target: $ARGUMENTS
 
 ## Instructions
 
-Parse the arguments to extract:
-- `--prompt`: the review prompt (required)
-- `--models`: comma-separated model tiers (default: `sonnet,opus`)
-- `--score`: enable judge scoring pipeline (optional flag)
-- `--judge`: judge model tier (default: `opus`)
-- `--task-type`: scoring category — `code`, `review`, `analysis`, `text`, `general` (default: `general`)
-- `--timeout`: per-model timeout in seconds (default: 600)
-- `--output`: output directory (default: auto-generated tmp dir)
-
 1. Parse the user's arguments. Common forms:
 
    ```bash
    /cross-review "review this PR diff" --models sonnet,opus
    /cross-review "audit this code" --models sonnet,gemini-pro,gpt-4.1 --score
-   /cross-review "design this API" --score --judge opus --task-type analysis
+   /cross-review "design this API" --score --judge opus
    ```
 
 2. Run the cross-review:
@@ -41,11 +32,11 @@ Parse the arguments to extract:
      --models "sonnet,gemini-pro,gpt-4.1" \
      --score
 
-   # With custom judge model and task type
+   # With custom judge model
    ~/.aidevops/agents/scripts/compare-models-helper.sh cross-review \
      --prompt "your prompt here" \
      --models "sonnet,opus" \
-     --score --judge sonnet --task-type review
+     --score --judge sonnet
    ```
 
 3. Present the results:
@@ -55,18 +46,16 @@ Parse the arguments to extract:
    - Note any models that failed to respond
 
 4. If `--score` was used, scores are automatically:
-   - Recorded in the model-comparisons SQLite DB (`~/.aidevops/.agent-workspace/memory/model-comparisons.db`)
-   - Fed into the pattern tracker for data-driven model routing (`/route`, `/patterns`)
+   - Recorded in the model-comparisons SQLite DB
+   - Fed into the pattern tracker for model routing (`/route`, `/patterns`)
 
 ## Options
 
 | Option | Default | Description |
 |--------|---------|-------------|
-| `--prompt` | (required) | The review prompt |
 | `--models` | `sonnet,opus` | Comma-separated model tiers to compare |
 | `--score` | off | Auto-score outputs via judge model |
 | `--judge` | `opus` | Judge model tier (used with `--score`) |
-| `--task-type` | `general` | Scoring category: `code`, `review`, `analysis`, `text`, `general` |
 | `--timeout` | `600` | Seconds per model |
 | `--output` | auto | Directory for raw outputs |
 | `--workdir` | `pwd` | Working directory for model context |
@@ -77,14 +66,13 @@ Parse the arguments to extract:
 
 ## Scoring Criteria (judge model, 1-10 scale)
 
-| Criterion | Scale | Description |
-|-----------|-------|-------------|
-| Correctness | 1-10 | Factual accuracy and technical correctness |
-| Completeness | 1-10 | Coverage of all requirements and edge cases |
-| Quality | 1-10 | Code quality / writing quality |
-| Clarity | 1-10 | Clear explanation, good formatting, readability |
-| Adherence | 1-10 | Following instructions precisely, staying on-task |
-| Overall | 1-10 | Judge's holistic assessment |
+| Criterion | Description |
+|-----------|-------------|
+| correctness | Factual accuracy and technical correctness |
+| completeness | Coverage of all requirements and edge cases |
+| quality | Code quality, best practices, maintainability |
+| clarity | Clear explanation, good formatting, readability |
+| adherence | Following the original prompt instructions precisely |
 
 ## Examples
 
@@ -96,9 +84,6 @@ Parse the arguments to extract:
 /cross-review "Design a rate limiting strategy for a REST API" \
   --models sonnet,opus,pro --score
 
-# Custom judge model and task type
-/cross-review "Audit this architecture" --models "sonnet,opus" --score --judge opus --task-type analysis
-
 # Quick diff with custom timeout
 /cross-review "Summarize the key changes in this diff" --models haiku,sonnet --timeout 120
 
@@ -106,15 +91,6 @@ Parse the arguments to extract:
 /score-responses --leaderboard
 ```
 
-## Output
-
-- Per-model responses displayed inline
-- Diff summary (word counts, unified diff for 2-model comparisons)
-- Judge scores table (when `--score` is set)
-- Winner declaration with reasoning
-- Results saved to `~/.aidevops/.agent-workspace/tmp/cross-review-<timestamp>/`
-- Judge JSON saved to `<output_dir>/judge-scores.json`
-
 ## Related
 
 - `/compare-models` — Compare model capabilities and pricing (no live dispatch)
diff --git a/.agents/scripts/compare-models-helper.sh b/.agents/scripts/compare-models-helper.sh
index 0c4b2bf3ba..a72eade551 100755
--- a/.agents/scripts/compare-models-helper.sh
+++ b/.agents/scripts/compare-models-helper.sh
@@ -15,7 +15,7 @@
 #   capabilities  Show capability matrix
 #   providers     List supported providers
 #   discover      Detect available providers and models from local config
-#   cross-review  Dispatch same prompt to multiple models, diff results, optionally score via judge (t132.8, t1329)
+#   cross-review  Dispatch same prompt to multiple models, diff results (t132.8)
 #   help          Show this help
 #
 # Author: AI DevOps Framework
@@ -30,7 +30,7 @@ set -euo pipefail
 # Pattern Tracker Integration (t1098)
 # =============================================================================
 # Reads live success/failure data from the pattern tracker memory DB.
-# Same DB as pattern-tracker-helper.sh (archived) — no duplication of storage.
+# Same DB as pattern-tracker-helper.sh — no duplication of storage.
 
 readonly PATTERN_DB="${AIDEVOPS_MEMORY_DIR:-$HOME/.aidevops/.agent-workspace/memory}/memory.db"
 readonly -a PATTERN_VALID_MODELS=(haiku flash sonnet pro opus)
@@ -496,13 +496,13 @@ cmd_recommend() {
 		echo ""
 		echo "Pattern Tracker Insights:"
 		local pattern_lines
-		pattern_lines=$(get_all_tier_patterns)
+		pattern_lines=$(get_all_tier_patterns "")
 		if [[ -n "$pattern_lines" ]]; then
 			while IFS='|' read -r ptier prate psample; do
 				printf "  %-10s %d%% success (n=%d)\n" "$ptier:" "$prate" "$psample"
 			done <<<"$pattern_lines"
 		else
-			echo "  (no model-tagged patterns — record with /remember)"
+			echo "  (no model-tagged patterns — record with pattern-tracker-helper.sh)"
 		fi
 	fi
 
@@ -643,208 +643,281 @@ cmd_providers() {
 }
 
 # =============================================================================
-# Cross-Model Review (t132.8, t1329)
+# Cross-Model Review (t132.8)
 # Dispatch the same review prompt to multiple models, collect results, diff.
-# Optional --score flag dispatches outputs to a judge model for structured scoring.
 # =============================================================================
 
-#######################################
-# Resolve Anthropic API auth header (API key or OAuth)
-# Outputs: header string on stdout (e.g. "x-api-key: sk-...")
-# Returns: 0 on success, 1 if no auth available
-#######################################
-_resolve_cross_review_auth() {
-	local auth_file="${HOME}/.local/share/opencode/auth.json"
+# Judge scoring for cross-review (t1329)
+# Dispatches all model outputs to a judge model, parses structured JSON scores,
+# records results via cmd_score, and feeds into the pattern tracker.
+# Defined before cmd_cross_review (its caller) for readability.
+#
+# Args:
+#   $1 - original prompt
+#   $2 - models_str (comma-separated)
+#   $3 - output_dir
+#   $4 - judge_model tier
+#   $5+ - model_names array
+_cross_review_judge_score() {
+	local original_prompt="$1"
+	local models_str="$2"
+	local output_dir="$3"
+	local judge_model="$4"
+	shift 4
+	local -a model_names=("$@")
 
-	# Priority 1: environment variable
-	if [[ -n "${ANTHROPIC_API_KEY:-}" ]]; then
-		echo "x-api-key: ${ANTHROPIC_API_KEY}"
-		return 0
+	# Validate judge_model identifier (used in filenames and runner names)
+	if [[ ! "$judge_model" =~ ^[A-Za-z0-9._-]+$ ]]; then
+		print_error "Invalid judge model identifier: $judge_model"
+		return 1
 	fi
 
-	# Priority 2: OAuth token from OpenCode auth.json
-	if [[ -f "$auth_file" ]] && command -v jq &>/dev/null; then
-		local auth_type
-		auth_type=$(jq -r '.anthropic.type // empty' "$auth_file" 2>/dev/null)
-
-		if [[ "$auth_type" == "oauth" ]]; then
-			local access_token expires_at now_ms
-			access_token=$(jq -r '.anthropic.access // empty' "$auth_file" 2>/dev/null)
-			expires_at=$(jq -r '.anthropic.expires // 0' "$auth_file" 2>/dev/null)
-			now_ms=$(($(date +%s) * 1000))
-			if [[ -n "$access_token" && "$expires_at" -gt "$now_ms" ]]; then
-				echo "Authorization: Bearer ${access_token}"
-				return 0
-			fi
-		elif [[ "$auth_type" == "api" ]]; then
-			local api_key
-			api_key=$(jq -r '.anthropic.key // empty' "$auth_file" 2>/dev/null)
-			if [[ -n "$api_key" ]]; then
-				echo "x-api-key: ${api_key}"
-				return 0
-			fi
-		fi
+	local runner_helper="${SCRIPT_DIR}/runner-helper.sh"
+	if [[ ! -x "$runner_helper" ]]; then
+		print_warning "runner-helper.sh not found — skipping judge scoring"
+		return 0
 	fi
 
-	return 1
-}
-
-#######################################
-# Judge cross-review outputs via a judge model (t1329)
-# Reads model output files from output_dir, calls judge model API,
-# returns structured JSON scores to stdout.
-# Usage: _judge_cross_review <output_dir> <models_csv> <prompt> <judge_model> <task_type>
-# Returns: 0 on success (JSON on stdout), 1 on failure
-#######################################
-_judge_cross_review() {
-	local output_dir="$1"
-	local models_csv="$2"
-	local original_prompt="$3"
-	local judge_model="$4"
-	local task_type="${5:-general}"
-
-	# Require curl and jq
-	if ! command -v curl &>/dev/null || ! command -v jq &>/dev/null; then
-		print_error "Judge scoring requires curl and jq"
-		return 1
-	fi
+	echo "=== JUDGE SCORING (${judge_model}) ==="
+	echo ""
 
-	# Resolve auth
-	local auth_header
-	auth_header=$(_resolve_cross_review_auth) || {
-		print_error "Judge scoring requires Anthropic API key (ANTHROPIC_API_KEY or OpenCode OAuth)"
-		return 1
-	}
+	# Build judge prompt: include original prompt + all model responses
+	local judge_prompt
+	judge_prompt="You are a neutral judge evaluating AI model responses. Score each response on a 1-10 scale.
 
-	# Build model outputs section for the judge prompt
-	local outputs_text=""
-	local -a model_array=()
-	IFS=',' read -ra model_array <<<"$models_csv"
+ORIGINAL PROMPT:
+${original_prompt}
 
-	for model_tier in "${model_array[@]}"; do
-		model_tier=$(echo "$model_tier" | tr -d ' ')
+MODEL RESPONSES:
+"
+	# Bound per-model response size to keep judge payload within token limits
+	local max_chars_per_model=20000
+	local models_with_output=()
+	for model_tier in "${model_names[@]}"; do
 		local result_file="${output_dir}/${model_tier}.txt"
 		if [[ -f "$result_file" && -s "$result_file" ]]; then
 			local response_text
-			response_text=$(cat "$result_file")
-			outputs_text="${outputs_text}
-
+			response_text=$(head -c "$max_chars_per_model" "$result_file")
+			local file_size
+			file_size=$(wc -c <"$result_file" | tr -d ' ')
+			local truncated_marker=""
+			if [[ "$file_size" -gt "$max_chars_per_model" ]]; then
+				truncated_marker="
+[TRUNCATED — original ${file_size} chars, showing first ${max_chars_per_model}]"
+			fi
+			judge_prompt+="
 === MODEL: ${model_tier} ===
-${response_text}"
+${response_text}${truncated_marker}
+"
+			models_with_output+=("$model_tier")
 		fi
 	done
 
-	if [[ -z "$outputs_text" ]]; then
-		print_error "No model outputs found to judge in $output_dir"
-		return 1
+	if [[ ${#models_with_output[@]} -lt 2 ]]; then
+		print_warning "Not enough model outputs for judge scoring (need 2+)"
+		return 0
 	fi
 
-	# Build judge system prompt
-	local system_prompt
-	system_prompt='You are an expert AI model evaluator. You will be given a prompt and responses from multiple AI models. Score each model on five criteria (1-10 scale) and declare a winner.
-
-Scoring criteria:
-- correctness (1-10): Factual accuracy and technical correctness
-- completeness (1-10): Coverage of all requirements and edge cases
-- quality (1-10): Code quality, best practices, structure (or writing quality for non-code)
-- clarity (1-10): Clear explanation, good formatting, readability
-- adherence (1-10): Following instructions precisely, staying on-task
+	judge_prompt+="
+SCORING INSTRUCTIONS:
+Score each model on these criteria (1-10 scale):
+- correctness: Factual accuracy and technical correctness
+- completeness: Coverage of all requirements and edge cases
+- quality: Code quality, best practices, maintainability
+- clarity: Clear explanation, good formatting, readability
+- adherence: Following the original prompt instructions precisely
 
 Respond with ONLY a valid JSON object in this exact format (no markdown, no explanation):
 {
-  "scores": {
-    "<model_name>": {
-      "correctness": <1-10>,
-      "completeness": <1-10>,
-      "quality": <1-10>,
-      "clarity": <1-10>,
-      "adherence": <1-10>,
-      "overall": <1-10>,
-      "strengths": "<brief strengths>",
-      "weaknesses": "<brief weaknesses>"
+  \"task_type\": \"general\",
+  \"winner\": \"<model_tier_of_best_response>\",
+  \"reasoning\": \"<one sentence explaining the winner>\",
+  \"scores\": {
+    \"<model_tier>\": {
+      \"correctness\": <1-10>,
+      \"completeness\": <1-10>,
+      \"quality\": <1-10>,
+      \"clarity\": <1-10>,
+      \"adherence\": <1-10>
     }
-  },
-  "winner": "<model_name>",
-  "winner_reasoning": "<1-2 sentence rationale for winner>",
-  "task_type": "<code|analysis|text|review|general>"
-}'
-
-	local user_prompt
-	user_prompt="ORIGINAL PROMPT:
-${original_prompt}
+  }
+}"
 
-MODEL RESPONSES:
-${outputs_text}
+	# Sanitize judge_model before using in filenames/runner names
+	if [[ ! "$judge_model" =~ ^[A-Za-z0-9._-]+$ ]]; then
+		print_warning "Invalid judge model identifier: $judge_model — skipping scoring"
+		return 0
+	fi
 
-Score each model and declare a winner."
+	# Dispatch to judge model
+	local judge_runner="cross-review-judge-$$"
+	local judge_output_file="${output_dir}/judge-${judge_model}.json"
 
-	# Resolve judge model to full model ID
-	local judge_model_id
-	judge_model_id=$(resolve_model_tier "$judge_model" 2>/dev/null || echo "$judge_model")
+	echo "  Dispatching to judge (${judge_model})..."
 
-	# Anthropic Messages API expects raw model IDs without provider prefix
-	if [[ "$judge_model_id" == anthropic/* ]]; then
-		judge_model_id="${judge_model_id#anthropic/}"
-	elif [[ "$judge_model_id" == */* ]]; then
-		print_error "--judge must resolve to an Anthropic model when using Anthropic judge API (got: $judge_model_id)"
-		return 1
-	fi
+	local judge_err_log="${output_dir}/judge-errors.log"
+
+	"$runner_helper" create "$judge_runner" \
+		--model "$judge_model" \
+		--description "Cross-review judge" \
+		--workdir "$(pwd)" 2>>"$judge_err_log" || true
+
+	"$runner_helper" run "$judge_runner" "$judge_prompt" \
+		--model "$judge_model" \
+		--timeout "120" \
+		--format text >"$judge_output_file" 2>>"$judge_err_log" || true
+
+	"$runner_helper" destroy "$judge_runner" --force 2>>"$judge_err_log" || true
 
-	# Build API request body
-	local request_body
-	request_body=$(jq -n \
-		--arg model "$judge_model_id" \
-		--arg system "$system_prompt" \
-		--arg user "$user_prompt" \
-		'{model: $model, max_tokens: 1024, system: $system, messages: [{role: "user", content: $user}]}')
-
-	# Build curl config via process substitution to avoid exposing auth in ps
-	local header_name
-	header_name="${auth_header%%:*}"
-	local curl_config
-	curl_config="header = \"Content-Type: application/json\"
-header = \"anthropic-version: 2023-06-01\"
-header = \"${auth_header}\""
-	if [[ "$header_name" == "Authorization" ]]; then
-		curl_config="${curl_config}
-header = \"anthropic-beta: oauth-2025-04-20\""
+	if [[ ! -f "$judge_output_file" || ! -s "$judge_output_file" ]]; then
+		print_warning "Judge model returned no output — skipping scoring"
+		return 0
 	fi
 
-	# Call judge model API (headers via --config to keep secrets out of process list)
-	local response
-	response=$(curl -s --max-time 120 \
-		--config <(printf '%s\n' "$curl_config") \
-		-d "$request_body" \
-		"https://api.anthropic.com/v1/messages") || {
-		print_error "Judge API call failed (curl error)"
-		return 1
-	}
+	# Extract JSON from judge output (strip any surrounding text)
+	local judge_json
+	judge_json=$(grep -o '{.*}' "$judge_output_file" 2>>"$judge_err_log" | head -1 || true)
+	if [[ -z "$judge_json" ]]; then
+		# Try multiline JSON extraction via stdin (safe for paths with special chars)
+		judge_json=$(python3 -c "
+import sys, json, re
+text = sys.stdin.read()
+m = re.search(r'\{.*\}', text, re.DOTALL)
+if m:
+    try:
+        obj = json.loads(m.group())
+        print(json.dumps(obj))
+    except Exception:
+        pass
+" <"$judge_output_file" 2>>"$judge_err_log" || true)
+	fi
 
-	# Extract text content from response
-	local ai_text
-	ai_text=$(echo "$response" | jq -r '.content[]? | select(.type == "text") | .text')
+	if [[ -z "$judge_json" ]]; then
+		print_warning "Could not parse judge JSON output. Raw output saved to: $judge_output_file"
+		return 0
+	fi
 
-	if [[ -z "$ai_text" ]]; then
-		local api_error
-		api_error=$(echo "$response" | jq -r '.error.message // empty')
-		print_error "Judge API returned no content (error: ${api_error:-unknown})"
-		return 1
+	# Parse winner, task_type, and reasoning in a single Python call
+	local parsed_fields
+	parsed_fields=$(echo "$judge_json" | python3 -c "
+import sys, json
+d = json.load(sys.stdin)
+# Truncate reasoning to 500 chars and strip control characters
+r = d.get('reasoning', '')[:500]
+r = ''.join(c for c in r if c.isprintable() or c in (' ', '\t'))
+print(d.get('winner', ''))
+print(d.get('task_type', 'general'))
+print(r)
+" 2>>"$judge_err_log" || true)
+
+	local winner task_type reasoning
+	if [[ -n "$parsed_fields" ]]; then
+		winner=$(echo "$parsed_fields" | head -1)
+		task_type=$(echo "$parsed_fields" | sed -n '2p')
+		reasoning=$(echo "$parsed_fields" | sed -n '3p')
+	else
+		winner=""
+		task_type="general"
+		reasoning=""
 	fi
 
-	# Strip markdown code fences if present — extract full JSON object without truncation
-	local clean_json
-	clean_json=$(echo "$ai_text" | sed -n '/^{/,/^}/p')
-	if [[ -z "$clean_json" ]]; then
-		clean_json="$ai_text"
+	# Sanitize task_type: restrict to known allowlist
+	local -a valid_task_types=(general code review analysis debug refactor test docs security)
+	local task_type_valid=false
+	for vt in "${valid_task_types[@]}"; do
+		if [[ "$task_type" == "$vt" ]]; then
+			task_type_valid=true
+			break
+		fi
+	done
+	if [[ "$task_type_valid" != "true" ]]; then
+		task_type="general"
 	fi
 
-	# Validate it's parseable JSON
-	if ! echo "$clean_json" | jq . &>/dev/null; then
-		print_error "Judge returned invalid JSON"
-		return 1
+	# Sanitize winner: must be one of the models with output
+	local winner_valid=false
+	if [[ -n "$winner" ]]; then
+		for m in "${models_with_output[@]}"; do
+			if [[ "$winner" == "$m" ]]; then
+				winner_valid=true
+				break
+			fi
+		done
+		if [[ "$winner_valid" != "true" ]]; then
+			print_warning "Judge returned unknown winner '${winner}' — ignoring"
+			winner=""
+		fi
 	fi
 
-	echo "$clean_json"
+	echo "  Judge winner: ${winner:-unknown}"
+	[[ -n "$reasoning" ]] && echo "  Reasoning: ${reasoning}"
+	echo ""
+
+	# Helper: clamp a numeric value to integer in range 0-10
+	# Handles decimals correctly (e.g. 8.5 → 9 via rounding, not 85)
+	_clamp_score() {
+		local val="$1"
+		# Accept only valid numeric format (digits with optional single decimal point)
+		if [[ ! "$val" =~ ^[0-9]+(\.[0-9]+)?$ ]]; then
+			echo "0"
+			return 0
+		fi
+		# Round to nearest integer and clamp to 0-10
+		local int_val
+		int_val=$(printf '%.0f' "$val" 2>/dev/null || echo "0")
+		if [[ "$int_val" -gt 10 ]]; then
+			echo "10"
+		elif [[ "$int_val" -lt 0 ]]; then
+			echo "0"
+		else
+			echo "$int_val"
+		fi
+		return 0
+	}
+
+	# Build cmd_score arguments from judge JSON
+	local -a score_args=(
+		--task "$original_prompt"
+		--type "$task_type"
+		--evaluator "$judge_model"
+	)
+	[[ -n "$winner" ]] && score_args+=(--winner "$winner")
+
+	for model_tier in "${models_with_output[@]}"; do
+		# Extract all scores in a single Python call (avoids 5 subprocesses per model)
+		local scores_line
+		scores_line=$(echo "$judge_json" | python3 -c "
+import sys, json
+d = json.load(sys.stdin)
+s = d.get('scores', {}).get('${model_tier}', {})
+print(s.get('correctness', 0), s.get('completeness', 0), s.get('quality', 0), s.get('clarity', 0), s.get('adherence', 0))
+" 2>>"$judge_err_log" || echo "0 0 0 0 0")
+		local corr comp qual clar adhr
+		read -r corr comp qual clar adhr <<<"$scores_line"
+
+		# Clamp all scores to valid integer range 0-10
+		corr=$(_clamp_score "$corr")
+		comp=$(_clamp_score "$comp")
+		qual=$(_clamp_score "$qual")
+		clar=$(_clamp_score "$clar")
+		adhr=$(_clamp_score "$adhr")
+
+		score_args+=(
+			--model "$model_tier"
+			--correctness "$corr"
+			--completeness "$comp"
+			--quality "$qual"
+			--clarity "$clar"
+			--adherence "$adhr"
+		)
+	done
+
+	# Record scores via cmd_score (also syncs to pattern tracker)
+	cmd_score "${score_args[@]}"
+
+	echo "Judge scores recorded. Judge output: $judge_output_file"
+	echo ""
+
 	return 0
 }
 
@@ -852,13 +925,14 @@ header = \"anthropic-beta: oauth-2025-04-20\""
 # Cross-model review: dispatch same prompt to multiple models (t132.8, t1329)
 # Usage: compare-models-helper.sh cross-review --prompt "review this code" \
 #          --models "sonnet,opus,pro" [--workdir path] [--timeout N] [--output dir]
-#          [--score] [--judge opus] [--task-type code]
+#          [--score] [--judge <model>]
 # Dispatches via runner-helper.sh in parallel, collects outputs, produces summary.
-# With --score: dispatches outputs to judge model for structured scoring and DB recording.
+# With --score: feeds outputs to a judge model (default: opus) for structured scoring
+# and records results in the model-comparisons DB + pattern tracker.
 #######################################
 cmd_cross_review() {
 	local prompt="" models_str="" workdir="" review_timeout="600" output_dir=""
-	local do_score=0 judge_model="opus" task_type="general"
+	local score_flag=false judge_model="opus"
 
 	while [[ $# -gt 0 ]]; do
 		case "$1" in
@@ -903,7 +977,7 @@ cmd_cross_review() {
 			shift 2
 			;;
 		--score)
-			do_score=1
+			score_flag=true
 			shift
 			;;
 		--judge)
@@ -912,14 +986,11 @@ cmd_cross_review() {
 				return 1
 			}
 			judge_model="$2"
-			shift 2
-			;;
-		--task-type)
-			[[ $# -lt 2 ]] && {
-				print_error "--task-type requires a value"
+			# Validate judge model identifier (used in filenames)
+			if [[ ! "$judge_model" =~ ^[A-Za-z0-9._-]+$ ]]; then
+				print_error "Invalid judge model identifier: $judge_model (only alphanumeric, dots, hyphens, underscores)"
 				return 1
-			}
-			task_type="$2"
+			fi
 			shift 2
 			;;
 		*)
@@ -980,9 +1051,15 @@ cmd_cross_review() {
 	local -a model_names=()
 
 	for model_tier in "${model_array[@]}"; do
-		model_tier=$(echo "$model_tier" | tr -d ' ')
+		model_tier="${model_tier// /}"
 		[[ -z "$model_tier" ]] && continue
 
+		# Sanitize model identifier to prevent path traversal (reject chars outside safe set)
+		if [[ ! "$model_tier" =~ ^[A-Za-z0-9._-]+$ ]]; then
+			print_warning "Skipping invalid model identifier: $model_tier"
+			continue
+		fi
+
 		local runner_name="cross-review-${model_tier}-$$"
 		runner_names+=("$runner_name")
 		model_names+=("$model_tier")
@@ -993,27 +1070,34 @@ cmd_cross_review() {
 
 		echo "  Dispatching to ${model_tier} (${resolved_model})..."
 
-		# Create runner, dispatch, capture output
+		# Create runner, dispatch, capture output (errors logged per-model for debugging)
+		local model_err_log="${output_dir}/${model_tier}-errors.log"
 		(
+			local model_failed=0
+
 			"$runner_helper" create "$runner_name" \
 				--model "$model_tier" \
 				--description "Cross-review: $model_tier" \
-				--workdir "$workdir" 2>/dev/null || true
+				--workdir "$workdir" 2>>"$model_err_log" || model_failed=1
 
 			local result_file="${output_dir}/${model_tier}.txt"
 			"$runner_helper" run "$runner_name" "$prompt" \
 				--model "$model_tier" \
 				--timeout "$review_timeout" \
-				--format json 2>/dev/null >"${output_dir}/${model_tier}.json" || true
+				--format json 2>>"$model_err_log" >"${output_dir}/${model_tier}.json" || model_failed=1
 
 			# Extract text response from JSON
 			if [[ -f "${output_dir}/${model_tier}.json" ]]; then
 				jq -r '.parts[]? | select(.type == "text") | .text' \
-					"${output_dir}/${model_tier}.json" 2>/dev/null >"$result_file" || true
+					"${output_dir}/${model_tier}.json" 2>>"$model_err_log" >"$result_file" || model_failed=1
 			fi
 
-			# Clean up runner
-			"$runner_helper" destroy "$runner_name" --force 2>/dev/null || true
+			# Clean up runner (always attempt cleanup, even on failure)
+			"$runner_helper" destroy "$runner_name" --force 2>>"$model_err_log" || true
+
+			# Fail if no usable output was produced
+			[[ -s "$result_file" ]] || model_failed=1
+			exit "$model_failed"
 		) &
 		pids+=($!)
 	done
@@ -1024,7 +1108,8 @@ cmd_cross_review() {
 	local failed=0
 	for i in "${!pids[@]}"; do
 		if ! wait "${pids[$i]}" 2>/dev/null; then
-			echo "  ${model_names[$i]}: failed"
+			local err_log="${output_dir}/${model_names[$i]}-errors.log"
+			echo "  ${model_names[$i]}: failed (see ${err_log})"
 			failed=$((failed + 1))
 		else
 			echo "  ${model_names[$i]}: done"
@@ -1079,7 +1164,14 @@ cmd_cross_review() {
 		local file_b="${output_dir}/${model_names[1]}.txt"
 		if [[ -f "$file_a" && -f "$file_b" ]]; then
 			echo "Diff (${model_names[0]} vs ${model_names[1]}):"
-			diff --unified=3 "$file_a" "$file_b" || echo "  (files are identical or diff unavailable)"
+			# diff exits 1 when files differ — capture separately to avoid pipefail
+			local diff_output diff_status
+			diff_output=$(diff --unified=3 "$file_a" "$file_b" 2>/dev/null) && diff_status=$? || diff_status=$?
+			if [[ "$diff_status" -le 1 && -n "$diff_output" ]]; then
+				echo "$diff_output" | head -100
+			else
+				echo "  (files are identical or diff unavailable)"
+			fi
 			echo ""
 		fi
 	fi
@@ -1087,164 +1179,11 @@ cmd_cross_review() {
 	echo "Full results saved to: $output_dir"
 	echo ""
 
-	# ==========================================================================
-	# Judge scoring pipeline (t1329) — activated by --score flag
-	# ==========================================================================
-	if [[ "$do_score" -eq 1 ]]; then
-		echo "=== JUDGE SCORING ==="
-		echo ""
-		echo "Judge model: ${judge_model}"
-		echo "Task type:   ${task_type}"
-		echo ""
-		echo "Dispatching outputs to judge model..."
-
-		local judge_json
-		if ! judge_json=$(_judge_cross_review "$output_dir" "$models_str" "$prompt" "$judge_model" "$task_type"); then
-			print_error "Judge scoring failed — raw outputs still saved to $output_dir"
-			return 1
-		fi
-
-		# Save judge output
-		local judge_file="${output_dir}/judge-scores.json"
-		echo "$judge_json" >"$judge_file"
-
-		# Display structured scores
-		echo ""
-		echo "Judge Scores:"
-		echo "-------------"
-		printf "%-22s %6s %6s %6s %6s %7s\n" "Model" "Corr" "Comp" "Qual" "Clar" "Overall"
-		printf "%-22s %6s %6s %6s %6s %7s\n" "-----" "----" "----" "----" "----" "-------"
-
-		local winner=""
-		winner=$(echo "$judge_json" | jq -r '.winner // empty')
-		local winner_reasoning=""
-		winner_reasoning=$(echo "$judge_json" | jq -r '.winner_reasoning // empty')
-		local judge_task_type=""
-		judge_task_type=$(echo "$judge_json" | jq -r '.task_type // empty')
-		[[ -n "$judge_task_type" ]] && task_type="$judge_task_type"
-
-		# Build cmd_score args from judge JSON
-		local score_args=(--task "$prompt" --type "$task_type" --evaluator "$judge_model" --winner "$winner")
-
-		local -a scored_models=()
-		while IFS= read -r model_name; do
-			scored_models+=("$model_name")
-		done < <(echo "$judge_json" | jq -r '.scores | keys[]')
-
-		for model_name in "${scored_models[@]}"; do
-			# Consolidate jq calls into a single pass using --arg for safe variable passing
-			local m_corr m_comp m_qual m_clar m_adh m_overall m_str m_wea
-			read -r m_corr m_comp m_qual m_clar m_adh m_overall m_str m_wea <<<"$(echo "$judge_json" | jq -r --arg mn "$model_name" '
-				.scores[$mn] |
-				[
-					(.correctness // 0),
-					(.completeness // 0),
-					(.quality // 0),
-					(.clarity // 0),
-					(.adherence // 0),
-					(.overall // 0),
-					(.strengths // ""),
-					(.weaknesses // "")
-				] | @tsv
-			')"
-
-			local result_file="${output_dir}/${model_name}.txt"
-			printf "%-22s %6s %6s %6s %6s %7s\n" \
-				"$model_name" "$m_corr" "$m_comp" "$m_qual" "$m_clar" "$m_overall"
-
-			# Accumulate args for cmd_score (1-10 scale matches model-comparisons DB)
-			score_args+=(
-				--model "$model_name"
-				--correctness "$m_corr"
-				--completeness "$m_comp"
-				--quality "$m_qual"
-				--clarity "$m_clar"
-				--adherence "$m_adh"
-				--strengths "$m_str"
-				--weaknesses "$m_wea"
-				--response "${result_file:-}"
-			)
-		done
-
-		echo ""
-		if [[ -n "$winner" ]]; then
-			echo "  Winner: ${winner}"
-		fi
-		if [[ -n "$winner_reasoning" ]]; then
-			echo "  Reasoning: ${winner_reasoning}"
-		fi
-		echo ""
-
-		# Record scores in model-comparisons DB via cmd_score
-		echo "Recording scores in model-comparisons DB..."
-		if cmd_score "${score_args[@]}"; then
-			print_success "Scores recorded in model-comparisons DB"
-		else
-			print_warning "cmd_score recording failed (scores still in $judge_file)"
-		fi
-
-		# Feed winner/loser data into pattern tracker (archived — graceful fallback)
-		local pt_helper="${SCRIPT_DIR}/archived/pattern-tracker-helper.sh"
-		if [[ -x "$pt_helper" && -n "$winner" ]]; then
-			echo "Syncing results to pattern tracker..."
-			local winner_tier loser_args=() winner_overall_score=0
-			winner_tier=$(model_id_to_tier "$winner")
-			[[ -z "$winner_tier" ]] && winner_tier="$winner"
-
-			for model_name in "${scored_models[@]}"; do
-				# Extract scores in a single jq pass using --arg for safe variable passing
-				local m_ove raw_corr raw_comp raw_qual raw_clar
-				read -r m_ove raw_corr raw_comp raw_qual raw_clar <<<"$(echo "$judge_json" | jq -r --arg mn "$model_name" '
-					.scores[$mn] | [(.overall // 0), (.correctness // 0), (.completeness // 0), (.quality // 0), (.clarity // 0)] | @tsv
-				')"
-				local m_tier
-				m_tier=$(model_id_to_tier "$model_name")
-				[[ -z "$m_tier" ]] && m_tier="$model_name"
-
-				# Normalize 1-10 to 1-5 for pattern tracker
-				local norm_corr norm_comp norm_qual norm_clar
-				norm_corr=$(awk "BEGIN{v=int($raw_corr/2+0.5); if(v<1)v=1; if(v>5)v=5; print v}")
-				norm_comp=$(awk "BEGIN{v=int($raw_comp/2+0.5); if(v<1)v=1; if(v>5)v=5; print v}")
-				norm_qual=$(awk "BEGIN{v=int($raw_qual/2+0.5); if(v<1)v=1; if(v>5)v=5; print v}")
-				norm_clar=$(awk "BEGIN{v=int($raw_clar/2+0.5); if(v<1)v=1; if(v>5)v=5; print v}")
-
-				"$pt_helper" score \
-					--model "$m_tier" \
-					--task-type "$task_type" \
-					--correctness "$norm_corr" \
-					--completeness "$norm_comp" \
-					--code-quality "$norm_qual" \
-					--clarity "$norm_clar" \
-					--source "cross-review-judge" \
-					>/dev/null 2>&1 || true
-
-				if [[ "$model_name" == "$winner" ]]; then
-					winner_overall_score="$m_ove"
-				elif [[ "$model_name" != "$winner" ]]; then
-					loser_args+=(--loser "$m_tier")
-				fi
-			done
-
-			# Record A/B comparison in pattern tracker
-			if [[ -n "$winner_tier" && "${#loser_args[@]}" -gt 0 ]]; then
-				local winner_avg_norm
-				winner_avg_norm=$(awk "BEGIN{printf \"%.1f\", $winner_overall_score / 2}")
-				"$pt_helper" ab-compare \
-					--winner "$winner_tier" \
-					"${loser_args[@]}" \
-					--task-type "$task_type" \
-					--winner-score "$winner_avg_norm" \
-					--models-compared "${#scored_models[@]}" \
-					--source "cross-review-judge" \
-					>/dev/null 2>&1 || true
-			fi
-
-			print_success "Pattern tracker updated"
-		fi
-
-		echo ""
-		echo "Judge output saved to: $judge_file"
-		echo ""
+	# Judge scoring pipeline (t1329)
+	# When --score is set, dispatch all outputs to a judge model for structured scoring.
+	if [[ "$score_flag" == "true" ]]; then
+		_cross_review_judge_score \
+			"$prompt" "$models_str" "$output_dir" "$judge_model" "${model_names[@]}"
 	fi
 
 	return 0
@@ -1277,7 +1216,8 @@ cmd_patterns() {
 		echo "No pattern data available."
 		echo ""
 		echo "Record patterns to populate this view:"
-		echo "  /remember \"SUCCESS: code-review with sonnet — completed successfully\""
+		echo "  pattern-tracker-helper.sh record --outcome success --model sonnet --task-type code-review \\"
+		echo "    --description \"Completed code review successfully\""
 		echo ""
 		echo "The supervisor also records patterns automatically after each task."
 		return 0
@@ -1341,8 +1281,8 @@ cmd_patterns() {
 	fi
 
 	echo ""
-	echo "Data source: cross-session memory (memory.db)"
-	echo "Record more: /remember \"SUCCESS: <task-type> with <tier> — <description>\""
+	echo "Data source: pattern-tracker-helper.sh (memory.db)"
+	echo "Record more: pattern-tracker-helper.sh record --outcome success --model <tier> ..."
 	echo ""
 	return 0
 }
@@ -1405,16 +1345,11 @@ cmd_help() {
 	echo "    --prompt 'Audit the architecture of this project' \\"
 	echo "    --models 'opus,pro' --timeout 900"
 	echo "  compare-models-helper.sh cross-review \\"
-	echo "    --prompt 'Review this PR diff for bugs' \\"
-	echo "    --models 'sonnet,opus' --score"
+	echo "    --prompt 'Review this PR diff' --models 'sonnet,gemini-pro' \\"
+	echo "    --score                          # auto-score via judge model (default: opus)"
 	echo "  compare-models-helper.sh cross-review \\"
-	echo "    --prompt 'Review this PR diff for bugs' \\"
-	echo "    --models 'sonnet,opus,pro' --score --judge opus --task-type review"
-	echo ""
-	echo "Cross-review options:"
-	echo "  --score          Auto-score outputs via judge model (default: opus)"
-	echo "  --judge <model>  Judge model tier (default: opus)"
-	echo "  --task-type <t>  Task type for scoring: code|review|analysis|text|general"
+	echo "    --prompt 'Review this PR diff' --models 'sonnet,gemini-pro' \\"
+	echo "    --score --judge sonnet            # use sonnet as judge instead"
 	echo ""
 	echo "Data is embedded in this script. Last updated: 2025-02-08."
 	echo "For live pricing, use /compare-models (with web fetch)."
@@ -1956,14 +1891,37 @@ cmd_score() {
 		return 1
 	fi
 
-	# Insert comparison record
-	local comp_id
-	comp_id=$(sqlite3 "$RESULTS_DB" "INSERT INTO comparisons (task_description, task_type, evaluator_model, winner_model) VALUES ('$(echo "$task" | sed "s/'/''/g")', '$task_type', '$evaluator', '$winner'); SELECT last_insert_rowid();")
+	# Insert comparison record (escape all string values for SQL safety)
+	local comp_id safe_task safe_type safe_eval safe_winner
+	safe_task="${task//\'/\'\'}"
+	safe_type="${task_type//\'/\'\'}"
+	safe_eval="${evaluator//\'/\'\'}"
+	safe_winner="${winner//\'/\'\'}"
+	comp_id=$(sqlite3 "$RESULTS_DB" "INSERT INTO comparisons (task_description, task_type, evaluator_model, winner_model) VALUES ('${safe_task}', '${safe_type}', '${safe_eval}', '${safe_winner}'); SELECT last_insert_rowid();")
 
-	# Insert scores for each model
+	# Insert scores for each model (escape strings, validate numerics)
 	for entry in "${model_entries[@]}"; do
 		IFS='|' read -r m_id m_cor m_com m_qua m_cla m_adh m_ove m_lat m_tok m_str m_wea m_res <<<"$entry"
-		sqlite3 "$RESULTS_DB" "INSERT INTO comparison_scores (comparison_id, model_id, correctness, completeness, code_quality, clarity, adherence, overall, latency_ms, tokens_used, strengths, weaknesses, response_file) VALUES ($comp_id, '$m_id', $m_cor, $m_com, $m_qua, $m_cla, $m_adh, $m_ove, $m_lat, $m_tok, '$(echo "$m_str" | sed "s/'/''/g")', '$(echo "$m_wea" | sed "s/'/''/g")', '$(echo "$m_res" | sed "s/'/''/g")');"
+
+		# Validate all numeric fields — reject non-integer values to prevent SQL injection
+		for n in m_cor m_com m_qua m_cla m_adh m_ove m_lat m_tok; do
+			if ! [[ "${!n}" =~ ^[0-9]+$ ]]; then
+				print_error "Invalid numeric value for ${n}: ${!n}"
+				return 1
+			fi
+		done
+		# Clamp score fields to valid 0-10 range
+		for s in m_cor m_com m_qua m_cla m_adh m_ove; do
+			if ((${!s} > 10)); then
+				printf -v "$s" "10"
+			fi
+		done
+
+		local safe_id="${m_id//\'/\'\'}"
+		local safe_str="${m_str//\'/\'\'}"
+		local safe_wea="${m_wea//\'/\'\'}"
+		local safe_res="${m_res//\'/\'\'}"
+		sqlite3 "$RESULTS_DB" "INSERT INTO comparison_scores (comparison_id, model_id, correctness, completeness, code_quality, clarity, adherence, overall, latency_ms, tokens_used, strengths, weaknesses, response_file) VALUES ($comp_id, '${safe_id}', $m_cor, $m_com, $m_qua, $m_cla, $m_adh, $m_ove, $m_lat, $m_tok, '${safe_str}', '${safe_wea}', '${safe_res}');"
 	done
 
 	print_success "Comparison #$comp_id recorded ($task_type: ${#model_entries[@]} models scored)"
@@ -1990,9 +1948,9 @@ cmd_score() {
 	fi
 	echo ""
 
-	# Sync to unified pattern tracker backbone (t1094) — archived, graceful fallback
+	# Sync to unified pattern tracker backbone (t1094)
 	# Scores are 1-10 here; normalize to 1-5 for pattern tracker compatibility.
-	local pt_helper="${SCRIPT_DIR}/archived/pattern-tracker-helper.sh"
+	local pt_helper="${SCRIPT_DIR}/pattern-tracker-helper.sh"
 	if [[ -x "$pt_helper" ]]; then
 		local winner_tier=""
 		local loser_args=()
@@ -2077,15 +2035,25 @@ cmd_results() {
 		esac
 	done
 
+	# Validate limit is numeric (used in SQL LIMIT clause)
+	if ! [[ "$limit" =~ ^[0-9]+$ ]]; then
+		print_error "Invalid --limit value: $limit (must be a positive integer)"
+		return 1
+	fi
+
+	# Escape string values for SQL safety (prevent injection via --model/--type args)
+	local safe_model_filter="${model_filter//\'/\'\'}"
+	local safe_type_filter="${type_filter//\'/\'\'}"
+
 	local where_clause=""
-	if [[ -n "$model_filter" ]]; then
-		where_clause="WHERE cs.model_id LIKE '%${model_filter}%'"
+	if [[ -n "$safe_model_filter" ]]; then
+		where_clause="WHERE cs.model_id LIKE '%${safe_model_filter}%'"
 	fi
-	if [[ -n "$type_filter" ]]; then
+	if [[ -n "$safe_type_filter" ]]; then
 		if [[ -n "$where_clause" ]]; then
-			where_clause="$where_clause AND c.task_type = '$type_filter'"
+			where_clause="$where_clause AND c.task_type = '${safe_type_filter}'"
 		else
-			where_clause="WHERE c.task_type = '$type_filter'"
+			where_clause="WHERE c.task_type = '${safe_type_filter}'"
 		fi
 	fi
 
diff --git a/.agents/scripts/matterbridge-helper.sh b/.agents/scripts/matterbridge-helper.sh
index 43ef996ae8..c996d5db7b 100755
--- a/.agents/scripts/matterbridge-helper.sh
+++ b/.agents/scripts/matterbridge-helper.sh
@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # matterbridge-helper.sh — Manage Matterbridge multi-platform chat bridge
-# Usage: matterbridge-helper.sh [setup|start|stop|status|logs|validate|update|simplex-bridge]
+# Usage: matterbridge-helper.sh [setup|start|stop|status|logs|validate|update]
 set -euo pipefail
 
 BINARY_PATH="/usr/local/bin/matterbridge"
@@ -9,9 +9,6 @@ DATA_DIR="$HOME/.aidevops/.agent-workspace/matterbridge"
 PID_FILE="$DATA_DIR/matterbridge.pid"
 LOG_FILE="$DATA_DIR/matterbridge.log"
 LATEST_RELEASE_URL="https://api.github.com/repos/42wim/matterbridge/releases/latest"
-SIMPLEX_BRIDGE_DIR="$DATA_DIR/simplex-bridge"
-SIMPLEX_COMPOSE_FILE="${SIMPLEX_COMPOSE_FILE:-}"
-AGENTS_DIR="${AGENTS_DIR:-$HOME/.aidevops/agents}"
 
 # ── helpers ──────────────────────────────────────────────────────────────────
 
@@ -33,9 +30,20 @@ ensure_dirs() {
 }
 
 get_latest_version() {
-	local version
-	version=$(curl -fsSL "$LATEST_RELEASE_URL" 2>/dev/null | grep '"tag_name"' | head -1 | sed 's/.*"v\([^"]*\)".*/\1/')
-	echo "${version:-1.26.0}"
+	local version curl_output curl_status
+	curl_output=$(curl -fsSL "$LATEST_RELEASE_URL" 2>&1) && curl_status=0 || curl_status=$?
+	if [[ "$curl_status" -ne 0 ]]; then
+		log "WARNING: Could not fetch latest version (curl exit $curl_status) — using fallback"
+		echo "1.26.0"
+		return 0
+	fi
+	version=$(echo "$curl_output" | grep '"tag_name"' | head -1 | sed 's/.*"v\([^"]*\)".*/\1/')
+	if [[ -z "$version" ]]; then
+		log "WARNING: Could not parse version from API response — using fallback"
+		echo "1.26.0"
+		return 0
+	fi
+	echo "$version"
 	return 0
 }
 
@@ -52,7 +60,14 @@ detect_os_arch() {
 
 	case "$os" in
 	linux) echo "linux-${arch}" ;;
-	darwin) echo "darwin-amd64" ;;
+	darwin)
+		# Matterbridge releases use darwin-arm64 and darwin-64bit (not darwin-amd64)
+		if [[ "$arch" == "arm64" ]]; then
+			echo "darwin-arm64"
+		else
+			echo "darwin-64bit"
+		fi
+		;;
 	*) echo "linux-${arch}" ;;
 	esac
 	return 0
@@ -152,11 +167,40 @@ cmd_validate() {
 		return 1
 	fi
 
-	# Basic TOML syntax check via matterbridge dry-run
-	# matterbridge exits non-zero if config is invalid
+	# Check binary exists and is executable
 	log "Validating config: $config_path"
-	if timeout 5 "$BINARY_PATH" -conf "$config_path" -version >/dev/null 2>&1; then
-		log "Binary OK"
+	if [ ! -x "$BINARY_PATH" ]; then
+		die "Binary not executable: $BINARY_PATH"
+		return 1
+	fi
+	log "Binary OK: $BINARY_PATH"
+
+	# Attempt to parse config (matterbridge will fail fast on invalid TOML)
+	# Use gtimeout on macOS if timeout is unavailable
+	local timeout_cmd="timeout"
+	if ! command -v timeout >/dev/null 2>&1; then
+		if command -v gtimeout >/dev/null 2>&1; then
+			timeout_cmd="gtimeout"
+		else
+			log "WARNING: timeout/gtimeout not found — skipping config parse check"
+			return 0
+		fi
+	fi
+
+	local parse_output parse_status
+	parse_output=$("$timeout_cmd" 5 "$BINARY_PATH" -conf "$config_path" 2>&1) && parse_status=$? || parse_status=$?
+
+	if [[ "$parse_status" -eq 124 ]]; then
+		# timeout exit code 124 = process timed out (likely hung on credentials)
+		log "Config parse: process timed out (expected if credentials are not configured)"
+	elif [[ "$parse_status" -ne 0 ]]; then
+		# Non-zero exit — check if it's a config parse error
+		if echo "$parse_output" | grep -qi "toml\|parse\|syntax"; then
+			die "Config parse error: $parse_output"
+			return 1
+		fi
+		# Other non-zero exits are expected (e.g., missing credentials)
+		log "Config parse: binary exited $parse_status (expected if credentials are not configured)"
 	fi
 
 	# Check for required sections
@@ -216,18 +260,24 @@ cmd_stop() {
 	local pid
 	pid="$(cat "$PID_FILE")"
 	log "Stopping (PID: $pid)..."
-	kill "$pid" 2>/dev/null || true
+	# kill may fail if process exited between is_running check and here (race condition)
+	local kill_err
+	kill_err=$(kill "$pid" 2>&1) || {
+		if [[ -n "$kill_err" ]]; then
+			log "WARNING: kill failed: $kill_err"
+		fi
+	}
 
-	local timeout=10
+	local stop_timeout=10
 	local count=0
-	while is_running && [ $count -lt $timeout ]; do
+	while is_running && [ $count -lt $stop_timeout ]; do
 		sleep 1
 		count=$((count + 1))
 	done
 
 	if is_running; then
 		log "Force killing..."
-		kill -9 "$pid" 2>/dev/null || true
+		kill -9 "$pid" 2>&1 | while IFS= read -r line; do log "WARNING: $line"; done || true
 	fi
 
 	rm -f "$PID_FILE"
@@ -251,12 +301,27 @@ cmd_status() {
 cmd_logs() {
 	local follow=false
 	local tail_lines=50
-	local arg="${1:-}"
 
-	case "$arg" in
-	--follow | -f) follow=true ;;
-	--tail) tail_lines="${2:-50}" ;;
-	esac
+	while [[ $# -gt 0 ]]; do
+		case "$1" in
+		--follow | -f)
+			follow=true
+			shift
+			;;
+		--tail)
+			if [[ -n "${2:-}" && "${2:-}" != -* ]]; then
+				tail_lines="$2"
+				shift 2
+			else
+				tail_lines=50
+				shift
+			fi
+			;;
+		*)
+			shift
+			;;
+		esac
+	done
 
 	if [ ! -f "$LOG_FILE" ]; then
 		log "No log file found: $LOG_FILE"
@@ -273,8 +338,17 @@ cmd_logs() {
 }
 
 cmd_update() {
+	if [ ! -f "$BINARY_PATH" ]; then
+		die "Binary not found: $BINARY_PATH. Run: matterbridge-helper.sh setup"
+		return 1
+	fi
+	# Ensure DATA_DIR exists before writing temp files (LOG_FILE lives under DATA_DIR)
+	ensure_dirs
+
 	local current_version new_version
-	current_version="$("$BINARY_PATH" -version 2>&1 | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1 || echo "unknown")"
+	local version_err_file="${LOG_FILE}.version-err"
+	current_version="$("$BINARY_PATH" -version 2>"$version_err_file" | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1 || echo "unknown")"
+	rm -f "$version_err_file"
 	new_version="$(get_latest_version)"
 
 	if [ "$current_version" = "$new_version" ]; then
@@ -308,247 +382,6 @@ cmd_update() {
 	return 0
 }
 
-# ── simplex bridge (Docker Compose) ──────────────────────────────────────────
-
-_find_compose_file() {
-	# Priority: env var > simplex-bridge dir > agents configs dir
-	if [ -n "$SIMPLEX_COMPOSE_FILE" ] && [ -f "$SIMPLEX_COMPOSE_FILE" ]; then
-		echo "$SIMPLEX_COMPOSE_FILE"
-		return 0
-	fi
-	if [ -f "$SIMPLEX_BRIDGE_DIR/docker-compose.yml" ]; then
-		echo "$SIMPLEX_BRIDGE_DIR/docker-compose.yml"
-		return 0
-	fi
-	if [ -f "$AGENTS_DIR/configs/matterbridge-simplex-compose.yml" ]; then
-		echo "$AGENTS_DIR/configs/matterbridge-simplex-compose.yml"
-		return 0
-	fi
-	# Check repo-local path (for development)
-	local script_dir
-	script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-	local repo_compose="${script_dir}/../configs/matterbridge-simplex-compose.yml"
-	if [ -f "$repo_compose" ]; then
-		echo "$repo_compose"
-		return 0
-	fi
-	echo ""
-	return 1
-}
-
-_ensure_simplex_bridge_dir() {
-	mkdir -p "$SIMPLEX_BRIDGE_DIR"
-	mkdir -p "$SIMPLEX_BRIDGE_DIR/data/simplex"
-	return 0
-}
-
-cmd_simplex_bridge() {
-	local action="${1:-help}"
-	shift || true
-
-	case "$action" in
-	up | start)
-		_simplex_bridge_up "$@"
-		;;
-	down | stop)
-		_simplex_bridge_down "$@"
-		;;
-	status | ps)
-		_simplex_bridge_status
-		;;
-	logs)
-		_simplex_bridge_logs "$@"
-		;;
-	init | setup)
-		_simplex_bridge_init "$@"
-		;;
-	help | --help | -h)
-		_simplex_bridge_help
-		;;
-	*)
-		log "Unknown simplex-bridge action: $action"
-		_simplex_bridge_help
-		return 1
-		;;
-	esac
-
-	return 0
-}
-
-_simplex_bridge_up() {
-	local compose_file
-	compose_file="$(_find_compose_file)" || {
-		die "No compose file found. Run: matterbridge-helper.sh simplex-bridge init"
-		return 1
-	}
-
-	if ! command -v docker >/dev/null 2>&1; then
-		die "Docker not found. Install Docker or OrbStack first."
-		return 1
-	fi
-
-	# Check for matterbridge.toml in the compose file directory
-	local compose_dir
-	compose_dir="$(dirname "$compose_file")"
-	if [ ! -f "$compose_dir/matterbridge.toml" ] && [ ! -f "$SIMPLEX_BRIDGE_DIR/matterbridge.toml" ]; then
-		log "WARNING: No matterbridge.toml found. The bridge will not connect to any platforms."
-		log "Copy the template: cp configs/matterbridge-simplex.toml.example matterbridge.toml"
-	fi
-
-	# Check for SimpleX database
-	if [ ! -d "$SIMPLEX_BRIDGE_DIR/data/simplex" ] || [ -z "$(ls -A "$SIMPLEX_BRIDGE_DIR/data/simplex" 2>/dev/null)" ]; then
-		log "WARNING: No SimpleX database found in $SIMPLEX_BRIDGE_DIR/data/simplex/"
-		log "Run simplex-chat CLI first to create a profile, then copy database files."
-		log "See: matterbridge-helper.sh simplex-bridge init"
-	fi
-
-	log "Starting SimpleX bridge stack..."
-	docker compose -f "$compose_file" up --build -d "$@"
-	log "SimpleX bridge started. Check status: matterbridge-helper.sh simplex-bridge status"
-	return 0
-}
-
-_simplex_bridge_down() {
-	local compose_file
-	compose_file="$(_find_compose_file)" || {
-		die "No compose file found."
-		return 1
-	}
-
-	log "Stopping SimpleX bridge stack..."
-	docker compose -f "$compose_file" down "$@"
-	log "SimpleX bridge stopped."
-	return 0
-}
-
-_simplex_bridge_status() {
-	local compose_file
-	compose_file="$(_find_compose_file)" || {
-		log "No compose file found. SimpleX bridge not configured."
-		return 0
-	}
-
-	log "Compose file: $compose_file"
-	log "Data dir: $SIMPLEX_BRIDGE_DIR"
-
-	if command -v docker >/dev/null 2>&1; then
-		docker compose -f "$compose_file" ps 2>/dev/null || log "No containers running"
-	else
-		log "Docker not available"
-	fi
-
-	# Check for SimpleX database
-	if [ -d "$SIMPLEX_BRIDGE_DIR/data/simplex" ]; then
-		local db_count
-		db_count="$(find "$SIMPLEX_BRIDGE_DIR/data/simplex" -name 'simplex_v1_*' 2>/dev/null | wc -l | tr -d ' ')"
-		log "SimpleX database files: $db_count"
-	else
-		log "SimpleX database: not found"
-	fi
-
-	return 0
-}
-
-_simplex_bridge_logs() {
-	local compose_file
-	compose_file="$(_find_compose_file)" || {
-		die "No compose file found."
-		return 1
-	}
-
-	docker compose -f "$compose_file" logs "$@"
-	return 0
-}
-
-_simplex_bridge_init() {
-	_ensure_simplex_bridge_dir
-
-	# Copy compose file to working directory if not present
-	local compose_file
-	compose_file="$(_find_compose_file)" || true
-
-	if [ -z "$compose_file" ] || [ ! -f "$SIMPLEX_BRIDGE_DIR/docker-compose.yml" ]; then
-		local template=""
-		# Find template from agents dir or repo
-		if [ -f "$AGENTS_DIR/configs/matterbridge-simplex-compose.yml" ]; then
-			template="$AGENTS_DIR/configs/matterbridge-simplex-compose.yml"
-		else
-			local script_dir
-			script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-			local repo_template="${script_dir}/../configs/matterbridge-simplex-compose.yml"
-			if [ -f "$repo_template" ]; then
-				template="$repo_template"
-			fi
-		fi
-
-		if [ -n "$template" ]; then
-			cp "$template" "$SIMPLEX_BRIDGE_DIR/docker-compose.yml"
-			log "Copied compose template to $SIMPLEX_BRIDGE_DIR/docker-compose.yml"
-		else
-			die "No compose template found. Ensure aidevops is installed."
-			return 1
-		fi
-	else
-		log "Compose file already exists: $SIMPLEX_BRIDGE_DIR/docker-compose.yml"
-	fi
-
-	# Copy config template if not present
-	if [ ! -f "$SIMPLEX_BRIDGE_DIR/matterbridge.toml" ]; then
-		local config_template=""
-		if [ -f "$AGENTS_DIR/configs/matterbridge-simplex.toml.example" ]; then
-			config_template="$AGENTS_DIR/configs/matterbridge-simplex.toml.example"
-		else
-			local script_dir
-			script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-			local repo_config="${script_dir}/../configs/matterbridge-simplex.toml.example"
-			if [ -f "$repo_config" ]; then
-				config_template="$repo_config"
-			fi
-		fi
-
-		if [ -n "$config_template" ]; then
-			cp "$config_template" "$SIMPLEX_BRIDGE_DIR/matterbridge.toml"
-			chmod 600 "$SIMPLEX_BRIDGE_DIR/matterbridge.toml"
-			log "Copied config template to $SIMPLEX_BRIDGE_DIR/matterbridge.toml"
-		else
-			log "WARNING: No config template found. Create matterbridge.toml manually."
-		fi
-	else
-		log "Config already exists: $SIMPLEX_BRIDGE_DIR/matterbridge.toml"
-	fi
-
-	log ""
-	log "Next steps:"
-	log "  1. Run simplex-chat CLI to create a profile and join/create the chat to bridge"
-	log "  2. Copy database: cp ~/.simplex/simplex_v1_* $SIMPLEX_BRIDGE_DIR/data/simplex/"
-	log "  3. Set permissions: chmod -R 777 $SIMPLEX_BRIDGE_DIR/data/"
-	log "  4. Get chat ID: simplex-chat -e '/i #group_name'"
-	log "  5. Edit $SIMPLEX_BRIDGE_DIR/matterbridge.toml with platform credentials"
-	log "  6. Start: matterbridge-helper.sh simplex-bridge up"
-	return 0
-}
-
-_simplex_bridge_help() {
-	cat <<'HELP'
-matterbridge-helper.sh simplex-bridge — Manage SimpleX-Matterbridge Docker stack
-
-Actions:
-  init               Set up working directory with compose + config templates
-  up                 Start the 3-container stack (simplex, matterbridge, adapter)
-  down               Stop and remove containers
-  status             Show container status and database info
-  logs [--follow]    Show container logs (pass -f for follow)
-
-Environment:
-  SIMPLEX_COMPOSE_FILE   Override compose file path
-  SIMPLEX_CHAT_ID        SimpleX chat ID to bridge (default: 1)
-  SIMPLEX_CHAT_TYPE      Chat type: contact or group (default: group)
-
-Docs: .agents/services/communications/matterbridge.md
-HELP
-	return 0
-}
-
 cmd_help() {
 	cat <<'HELP'
 matterbridge-helper.sh — Manage Matterbridge multi-platform chat bridge
@@ -561,7 +394,6 @@ Commands:
   status             Show running status
   logs [--follow]    Show/follow log output
   update             Update to latest release
-  simplex-bridge     Manage SimpleX bridge Docker stack (init|up|down|status|logs)
 
 Config: ~/.config/aidevops/matterbridge.toml (override: MATTERBRIDGE_CONFIG)
 Docs:   .agents/services/communications/matterbridge.md
@@ -583,7 +415,6 @@ main() {
 	status) cmd_status "$@" ;;
 	logs) cmd_logs "$@" ;;
 	update) cmd_update "$@" ;;
-	simplex-bridge) cmd_simplex_bridge "$@" ;;
 	help | --help | -h) cmd_help ;;
 	*)
 		echo "Unknown command: $cmd" >&2
diff --git a/.agents/services/communications/matterbridge.md b/.agents/services/communications/matterbridge.md
index e64feeb5af..f9c938725b 100644
--- a/.agents/services/communications/matterbridge.md
+++ b/.agents/services/communications/matterbridge.md
@@ -20,7 +20,7 @@ tools:
 
 - **Repo**: [github.com/42wim/matterbridge](https://github.com/42wim/matterbridge) (7.4K stars, Apache-2.0, Go)
 - **Version**: v1.26.0 (stable)
-- **Script**: `matterbridge-helper.sh [setup|start|stop|status|logs|validate|simplex-bridge]`
+- **Script**: `matterbridge-helper.sh [setup|start|stop|status|logs|validate]`
 - **Config**: `~/.config/aidevops/matterbridge.toml` (600 permissions)
 - **Data**: `~/.aidevops/.agent-workspace/matterbridge/`
 - **Requires**: Go 1.18+ (build) or pre-compiled binary
@@ -62,7 +62,7 @@ matterbridge-helper.sh start --daemon
 
 ### 3rd Party via Matterbridge API
 
-- **SimpleX**: [matterbridge-simplex](https://github.com/UnkwUsr/matterbridge-simplex) adapter (MIT, Node.js) — routes via SimpleX CLI WebSocket API
+- **SimpleX**: [matterbridge-simplex](https://github.com/simplex-chat/matterbridge-simplex) adapter — routes via SimpleX CLI
 - **Delta Chat**: matterdelta
 - **Minecraft**: mattercraft, MatterBukkit
 
@@ -76,8 +76,13 @@ curl -L https://github.com/42wim/matterbridge/releases/latest/download/matterbri
   -o /usr/local/bin/matterbridge
 chmod +x /usr/local/bin/matterbridge
 
-# macOS
-curl -L https://github.com/42wim/matterbridge/releases/latest/download/matterbridge-1.26.0-darwin-amd64 \
+# macOS (Intel)
+curl -L https://github.com/42wim/matterbridge/releases/latest/download/matterbridge-1.26.0-darwin-64bit \
+  -o /usr/local/bin/matterbridge
+chmod +x /usr/local/bin/matterbridge
+
+# macOS (Apple Silicon)
+curl -L https://github.com/42wim/matterbridge/releases/latest/download/matterbridge-1.26.0-darwin-arm64 \
   -o /usr/local/bin/matterbridge
 chmod +x /usr/local/bin/matterbridge
 
@@ -299,135 +304,34 @@ enable=true
   RemoteNickFormat="[{PROTOCOL}] <{NICK}> "
 ```
 
-### SimpleX via matterbridge-simplex Adapter
-
-SimpleX is not natively supported by Matterbridge. The [matterbridge-simplex](https://github.com/UnkwUsr/matterbridge-simplex) adapter (MIT, Node.js) bridges SimpleX Chat to Matterbridge's HTTP API, enabling SimpleX to connect to all 40+ platforms.
-
-#### Architecture
-
-```text
-SimpleX CLI (WebSocket :5225)
-    |
-    | WebSocket JSON API (localhost)
-    |
-matterbridge-simplex (Node.js adapter)
-    |
-    | HTTP REST API (localhost:4242)
-    |
-Matterbridge (Go binary)
-    |
-    |--- Matrix rooms
-    |--- Telegram groups
-    |--- Discord channels
-    |--- Slack workspaces
-    |--- IRC channels
-    |--- 40+ other platforms
-```
-
-**Message flow (SimpleX -> other platforms)**:
-
-1. User sends message in SimpleX Chat (mobile/desktop/CLI)
-2. SimpleX CLI receives via SMP protocol, emits `newChatItems` event on WebSocket
-3. matterbridge-simplex adapter reads event, extracts text and sender
-4. Adapter POSTs message to Matterbridge HTTP API (`/api/message`)
-5. Matterbridge routes to all configured gateway destinations
-
-**Message flow (other platforms -> SimpleX)**:
-
-1. User sends message on Matrix/Telegram/Discord/etc.
-2. Matterbridge receives via platform SDK, buffers in API endpoint
-3. matterbridge-simplex adapter polls `/api/messages` (1s interval)
-4. Adapter sends message to SimpleX CLI via `apiSendTextMessage` WebSocket command
-5. SimpleX CLI delivers to the configured contact or group chat
-
-#### Features and Limitations
-
-| Feature | Status | Notes |
-|---------|--------|-------|
-| Text messages | Supported | Bidirectional |
-| Image previews | Supported | SimpleX -> other platforms (preview only, not full file) |
-| Full file transfer | Not yet | WIP in matterbridge-simplex |
-| `/hide` prefix | Supported | Messages starting with `/hide` are not bridged — SimpleX-only |
-| Contact chats | Supported | Bridge to a specific SimpleX contact |
-| Group chats | Supported | Bridge to a specific SimpleX group |
-| Multiple bridges | Supported | Run multiple adapter instances with different chat IDs |
-
-#### Quick Setup (Docker Compose)
-
-The recommended deployment uses Docker Compose with 3 containers. See `configs/matterbridge-simplex-compose.yml` for the full template.
-
-```bash
-# 1. Prepare SimpleX database
-#    Run simplex-chat CLI first to create a profile and join/create the chat to bridge
-simplex-chat
-#    Then move the database:
-mkdir -p data/simplex
-cp ~/.simplex/simplex_v1_* data/simplex/
-chmod -R 777 data/  # Required for Docker volume access
-
-# 2. Get the chat ID to bridge
-simplex-chat -e '/_get chats 1 pcc=off' \
-  | tail -n +2 \
-  | jq '.[].chatInfo | (.groupInfo // .contact) | {name: .localDisplayName, type: (if .groupId then "group" else "contact" end), id: .groupId // .contactId}'
-
-# 3. Configure matterbridge.toml (copy from configs/matterbridge-simplex.toml.example)
-cp configs/matterbridge-simplex.toml.example matterbridge.toml
-# Edit: add your platform credentials and channel IDs
-
-# 4. Deploy
-matterbridge-helper.sh simplex-bridge up
-
-# Or manually:
-docker compose -f configs/matterbridge-simplex-compose.yml up --build -d
-```
+### SimpleX via Adapter
 
-#### Quick Setup (Manual)
+SimpleX is not natively supported. Use [matterbridge-simplex](https://github.com/simplex-chat/matterbridge-simplex):
 
 ```bash
-# 1. Install SimpleX CLI
+# Install SimpleX CLI first
 curl -o- https://raw.githubusercontent.com/simplex-chat/simplex-chat/stable/install.sh | bash
 
-# 2. Clone matterbridge-simplex and build
-git clone https://github.com/UnkwUsr/matterbridge-simplex.git
-cd matterbridge-simplex
-git submodule update --init --recursive --depth 1
-( cd lib/simplex-chat-client-typescript/ && npm install && tsc )
+# Install matterbridge-simplex adapter
+go install github.com/simplex-chat/matterbridge-simplex@latest
 
-# 3. Start SimpleX CLI as WebSocket server
-simplex-chat -p 5225
-
-# 4. Start Matterbridge with API endpoint
-matterbridge -conf matterbridge.toml
-
-# 5. Start the adapter
-# Format: node main.js <MB_API_ADDR> <MB_GATEWAY> <SXC_WS_ADDR> <CHAT_ID> <CHAT_TYPE>
-node main.js 127.0.0.1:4242 gateway1 127.0.0.1:5225 1 group
+# Run adapter (exposes Matterbridge API endpoint)
+matterbridge-simplex --port 4242 --profile simplex-bridge
 ```
 
-#### Matterbridge Config for SimpleX
-
 ```toml
-# Matterbridge API endpoint — the adapter connects here
+# Matterbridge config: use API bridge to connect to adapter
 [api]
-  [api.myapi]
-  BindAddress="127.0.0.1:4242"
-  Buffer=1000
+  [api.simplex]
+  BindAddress="0.0.0.0:4243"
+  Token="your-api-token"
 
-# Add your destination platform(s)
-[matrix]
-  [matrix.home]
-  Server="https://matrix.example.com"
-  Login="bridgebot"
-  Password="YOUR_MATRIX_PASSWORD"
-  RemoteNickFormat="[SimpleX] <{NICK}> "
-
-# Gateway connecting SimpleX (via API) to Matrix
 [[gateway]]
 name="simplex-matrix"
 enable=true
 
   [[gateway.inout]]
-  account="api.myapi"
+  account="api.simplex"
   channel="api"
 
   [[gateway.inout]]
@@ -435,66 +339,7 @@ enable=true
   channel="#bridged:example.com"
 ```
 
-See `configs/matterbridge-simplex.toml.example` for a complete template with Matrix, Telegram, and Discord examples.
-
-#### Obtaining SimpleX Chat IDs
-
-SimpleX uses separate ID spaces for contacts and groups. You need both the ID and type.
-
-```bash
-# Get info for a specific chat
-simplex-chat -e '/i #group_name'    # Group
-simplex-chat -e '/i @contact_name'  # Contact
-
-# List all chats with IDs (requires jq)
-simplex-chat -e '/_get chats 1 pcc=off' \
-  | tail -n +2 \
-  | jq '.[].chatInfo | (.groupInfo // .contact) | {name: .localDisplayName, type: (if .groupId then "group" else "contact" end), id: .groupId // .contactId}'
-```
-
-**Note**: The default chat ID in the Docker Compose template is `4`. Check your actual chat ID and update `docker-compose.yml` if it differs.
-
-## Privacy Gradient
-
-Matterbridge enables a **privacy gradient** — users choose their preferred privacy level while staying in the same conversation.
-
-```text
-Maximum Privacy                                          Maximum Convenience
-|                                                                          |
-SimpleX Chat ──> Matrix (self-hosted) ──> Matrix (public) ──> Telegram/Discord
-No identifiers    Federated, E2E opt-in   Federated          Centralized
-No metadata       Server stores metadata  Server stores      Full metadata
-No phone/email    @user:server IDs        @user:server IDs   Phone/username
-```
-
-### How It Works
-
-1. **SimpleX users** get maximum privacy — no identifiers, no metadata, E2E encrypted
-2. **Matrix users** get federation and E2E encryption (when enabled), with `@user:server` identifiers
-3. **Telegram/Discord/Slack users** get convenience and existing ecosystem, with full platform metadata
-4. All users see the same messages, prefixed with `[Platform] <Username>` to identify origin
-5. SimpleX users can send `/hide` messages that are not bridged — visible only on SimpleX
-
-### Security Implications
-
-**E2E encryption is broken at bridge boundaries.** When bridging:
-
-- Messages are decrypted by the SimpleX CLI process
-- Passed in plaintext to the matterbridge-simplex adapter (localhost only)
-- Re-encrypted (or sent plaintext) to destination platform by Matterbridge
-- The bridge host has access to all message content in plaintext
-- Metadata (sender, timestamp) is visible to all bridged platforms
-
-**Mitigations**:
-
-- Run the entire bridge stack on a trusted, hardened host
-- Use `network_mode: host` in Docker (default) — all traffic stays on localhost
-- Use NetBird/WireGuard to restrict access to the bridge host
-- Store platform credentials in gopass: `aidevops secret set MATTERBRIDGE_MATRIX_TOKEN`
-- Config file must have 600 permissions: `chmod 600 matterbridge.toml`
-- Consider one-way bridges (`[[gateway.in]]`/`[[gateway.out]]`) to limit exposure
-
-See `tools/security/opsec.md` for full platform trust matrix and threat modeling.
+**Note**: SimpleX E2E encryption is broken at the bridge boundary. Messages entering the bridge are decrypted and re-encrypted for the destination platform. See `tools/security/opsec.md` for implications.
 
 ## Running
 
@@ -598,9 +443,22 @@ curl http://localhost:4242/api/messages \
 
 ## Security Considerations
 
-See [Privacy Gradient > Security Implications](#security-implications) for detailed analysis of E2E encryption at bridge boundaries.
+**E2E encryption is broken at bridge boundaries.** When bridging:
+
+- Messages are decrypted by Matterbridge process
+- Re-encrypted (or sent plaintext) to destination platform
+- The bridge host has access to all message content in plaintext
+- Metadata (sender, timestamp, channel) is visible to all bridged platforms
+
+**Mitigations**:
 
-**Summary**: E2E encryption is broken at bridge boundaries. The bridge host sees all messages in plaintext. Run on a trusted host, use localhost-only networking, store credentials in gopass, and set config file permissions to 600.
+- Run Matterbridge on a trusted, hardened host
+- Use NetBird/WireGuard to restrict access to the bridge host
+- Avoid bridging sensitive channels to unencrypted platforms (IRC, Slack, Discord)
+- Store credentials in gopass: `aidevops secret set MATTERBRIDGE_DISCORD_TOKEN`
+- Config file must have 600 permissions: `chmod 600 matterbridge.toml`
+
+See `tools/security/opsec.md` for full platform trust matrix and threat modeling.
 
 ## Related
 
@@ -608,5 +466,3 @@ See [Privacy Gradient > Security Implications](#security-implications) for detai
 - `services/communications/simplex.md` — SimpleX install, bot API, self-hosted servers
 - `tools/security/opsec.md` — Platform trust matrix, E2E status, metadata warnings
 - `tools/ai-assistants/headless-dispatch.md` — Headless dispatch patterns
-- `configs/matterbridge-simplex-compose.yml` — Docker Compose template for SimpleX bridge
-- `configs/matterbridge-simplex.toml.example` — Config template for SimpleX-Matrix bridging