diff --git a/.agents/AGENTS.md b/.agents/AGENTS.md
index d50d023bdf..e431e45a36 100644
--- a/.agents/AGENTS.md
+++ b/.agents/AGENTS.md
@@ -102,6 +102,8 @@ Use `/save-todo` after planning. Auto-detects complexity:
 
 **Dependencies**: `blocked-by:t001`, `blocks:t002`, `t001.1` (subtask)
 
+**GitHub issue sync**: Issue titles MUST be prefixed with `t{NNN}:` (followed by a space, e.g., `t{NNN}: <title>`). TODO.md tasks MUST include `ref:GH#{issue}`. Always maintain both directions. See `workflows/plans.md` "GitHub Issue Sync" section.
+
 **After ANY TODO/planning edit**: Commit and push immediately. Planning-only files (TODO.md, todo/) go directly to main — no branch, no PR. Mixed changes (planning + non-exception files) use a worktree. NEVER `git checkout -b` in the main repo. See `workflows/plans.md` "Commit and Push After TODO Changes" section.
 
 **Full docs**: `workflows/plans.md`, `tools/task-management/beads.md`
diff --git a/.agents/scripts/commands/log-issue-aidevops.md b/.agents/scripts/commands/log-issue-aidevops.md
index c7712e9904..2871bef304 100644
--- a/.agents/scripts/commands/log-issue-aidevops.md
+++ b/.agents/scripts/commands/log-issue-aidevops.md
@@ -129,11 +129,25 @@ Labels: bug (or enhancement, question, documentation)
 4. Cancel
 ```
 
+### Step 5b: Assign Task ID and Sync (Maintainers Only)
+
+> **Note**: This step is for aidevops maintainers with repository write access. External users reporting issues can skip to Step 6.
+
+Before creating the issue, assign a TODO.md task ID:
+
+1. Read TODO.md to find the highest existing t-number
+2. Assign the next sequential t-number (e.g., if highest is t148, use t149)
+3. Prefix the issue title with `t{NNN}:` (followed by a space)
+4. After issue creation, add the task to TODO.md with `ref:GH#{issue_number}`
+5. Commit and push TODO.md
+
+This ensures bidirectional sync between GitHub issues and TODO.md. See `workflows/plans.md` "GitHub Issue Sync" section.
+
 ### Step 6: Create the Issue
 
 ```bash
 gh issue create -R marcusquinn/aidevops \
-  --title "TITLE" \
+  --title "t{NNN}: TITLE" \
   --body "$(cat <<'EOF'
 BODY_CONTENT
 EOF
diff --git a/.agents/scripts/session-checkpoint-helper.sh b/.agents/scripts/session-checkpoint-helper.sh
new file mode 100755
index 0000000000..7b96f51a07
--- /dev/null
+++ b/.agents/scripts/session-checkpoint-helper.sh
@@ -0,0 +1,224 @@
+#!/usr/bin/env bash
+# session-checkpoint-helper.sh - Persist session state to survive context compaction
+# Part of aidevops framework: https://aidevops.sh
+#
+# Usage:
+#   session-checkpoint-helper.sh [command] [options]
+#
+# Commands:
+#   save              Save current session checkpoint
+#   load              Load and display current checkpoint
+#   clear             Remove checkpoint file
+#   status            Show checkpoint age and summary
+#   help              Show this help
+#
+# Options:
+#   --task <id>       Current task ID (e.g., t135.9)
+#   --next <ids>      Comma-separated next task IDs
+#   --worktree <path> Active worktree path
+#   --branch <name>   Active branch name
+#   --batch <name>    Batch/PR name
+#   --note <text>     Free-form context note
+#   --elapsed <mins>  Minutes elapsed in session
+#   --target <mins>   Target session duration in minutes
+#
+# The checkpoint file is written to:
+#   ~/.aidevops/.agent-workspace/tmp/session-checkpoint.md
+#
+# Design: AI agent writes checkpoint after each task completion and reads it
+# before starting the next task. Survives context compaction because state
+# is on disk, not in context window.
+
+set -euo pipefail
+
+readonly CHECKPOINT_DIR="${HOME}/.aidevops/.agent-workspace/tmp"
+readonly CHECKPOINT_FILE="${CHECKPOINT_DIR}/session-checkpoint.md"
+
+readonly GREEN='\033[0;32m'
+readonly YELLOW='\033[1;33m'
+readonly RED='\033[0;31m'
+readonly BOLD='\033[1m'
+readonly DIM='\033[2m'
+readonly NC='\033[0m'
+
+print_success() { printf "${GREEN}%s${NC}\n" "$1"; }
+print_warning() { printf "${YELLOW}%s${NC}\n" "$1"; }
+print_error() { printf "${RED}%s${NC}\n" "$1" >&2; }
+print_info() { printf "${DIM}%s${NC}\n" "$1"; }
+
+ensure_dir() {
+    if [[ ! -d "$CHECKPOINT_DIR" ]]; then
+        mkdir -p "$CHECKPOINT_DIR"
+    fi
+    return 0
+}
+
+cmd_save() {
+    local current_task=""
+    local next_tasks=""
+    local worktree_path=""
+    local branch_name=""
+    local batch_name=""
+    local note=""
+    local elapsed_mins=""
+    local target_mins=""
+
+    while [[ $# -gt 0 ]]; do
+        case "$1" in
+            --task) [[ $# -lt 2 ]] && { print_error "--task requires a value"; return 1; }; current_task="$2"; shift 2 ;;
+            --next) [[ $# -lt 2 ]] && { print_error "--next requires a value"; return 1; }; next_tasks="$2"; shift 2 ;;
+            --worktree) [[ $# -lt 2 ]] && { print_error "--worktree requires a value"; return 1; }; worktree_path="$2"; shift 2 ;;
+            --branch) [[ $# -lt 2 ]] && { print_error "--branch requires a value"; return 1; }; branch_name="$2"; shift 2 ;;
+            --batch) [[ $# -lt 2 ]] && { print_error "--batch requires a value"; return 1; }; batch_name="$2"; shift 2 ;;
+            --note) [[ $# -lt 2 ]] && { print_error "--note requires a value"; return 1; }; note="$2"; shift 2 ;;
+            --elapsed) [[ $# -lt 2 ]] && { print_error "--elapsed requires a value"; return 1; }; elapsed_mins="$2"; shift 2 ;;
+            --target) [[ $# -lt 2 ]] && { print_error "--target requires a value"; return 1; }; target_mins="$2"; shift 2 ;;
+            *) print_error "Unknown option: $1"; return 1 ;;
+        esac
+    done
+
+    ensure_dir
+
+    local timestamp
+    timestamp="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+
+    # Auto-detect git state if not provided
+    if [[ -z "$branch_name" ]]; then
+        branch_name="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "unknown")"
+    fi
+
+    # Build checkpoint file
+    cat > "$CHECKPOINT_FILE" <<EOF
+# Session Checkpoint
+
+Updated: ${timestamp}
+
+## Current State
+
+| Field | Value |
+|-------|-------|
+| Current Task | ${current_task:-none} |
+| Branch | ${branch_name} |
+| Worktree | ${worktree_path:-not set} |
+| Batch/PR | ${batch_name:-not set} |
+| Elapsed | ${elapsed_mins:-unknown} min |
+| Target | ${target_mins:-unknown} min |
+
+## Next Tasks
+
+${next_tasks:-No next tasks specified}
+
+## Context Note
+
+${note:-No additional context}
+
+## Git Status
+
+$(git status --short 2>/dev/null || echo "Not in a git repo")
+
+## Recent Commits (this branch)
+
+$(git log --oneline -5 2>/dev/null || echo "No commits")
+
+## Open Worktrees
+
+$(git worktree list 2>/dev/null || echo "No worktrees")
+EOF
+
+    print_success "Checkpoint saved: ${CHECKPOINT_FILE}"
+    print_info "Task: ${current_task:-none} | Branch: ${branch_name} | ${timestamp}"
+    return 0
+}
+
+cmd_load() {
+    if [[ ! -f "$CHECKPOINT_FILE" ]]; then
+        print_warning "No checkpoint found at ${CHECKPOINT_FILE}"
+        print_info "Run: session-checkpoint-helper.sh save --task <id> --next <ids>"
+        return 1
+    fi
+
+    cat "$CHECKPOINT_FILE"
+    return 0
+}
+
+cmd_clear() {
+    if [[ -f "$CHECKPOINT_FILE" ]]; then
+        rm "$CHECKPOINT_FILE"
+        print_success "Checkpoint cleared"
+    else
+        print_info "No checkpoint to clear"
+    fi
+    return 0
+}
+
+cmd_status() {
+    if [[ ! -f "$CHECKPOINT_FILE" ]]; then
+        print_warning "No active checkpoint"
+        return 1
+    fi
+
+    local file_age_seconds
+    local now
+    local file_mtime
+
+    now="$(date +%s)"
+    if [[ "$(uname)" == "Darwin" ]]; then
+        file_mtime="$(stat -f %m "$CHECKPOINT_FILE")"
+    else
+        file_mtime="$(stat -c %Y "$CHECKPOINT_FILE")"
+    fi
+    file_age_seconds=$(( now - file_mtime ))
+
+    local age_display
+    if [[ $file_age_seconds -lt 60 ]]; then
+        age_display="${file_age_seconds}s ago"
+    elif [[ $file_age_seconds -lt 3600 ]]; then
+        age_display="$(( file_age_seconds / 60 ))m ago"
+    else
+        age_display="$(( file_age_seconds / 3600 ))h $(( (file_age_seconds % 3600) / 60 ))m ago"
+    fi
+
+    # Extract key fields
+    local current_task
+    current_task="$(awk -F'|' '/Current Task/ {gsub(/^[ \t]+|[ \t]+$/, "", $3); print $3; exit}' "$CHECKPOINT_FILE" || echo "unknown")"
+    local branch
+    branch="$(awk -F'|' '/Branch/ {gsub(/^[ \t]+|[ \t]+$/, "", $3); print $3; exit}' "$CHECKPOINT_FILE" || echo "unknown")"
+
+    printf '%b\n' "${BOLD}Checkpoint Status${NC}"
+    printf "  Age:    %s\n" "$age_display"
+    printf "  Task:   %s\n" "$current_task"
+    printf "  Branch: %s\n" "$branch"
+    printf "  File:   %s\n" "$CHECKPOINT_FILE"
+
+    if [[ $file_age_seconds -gt 1800 ]]; then
+        print_warning "  Warning: Checkpoint is stale (>30min). Consider updating."
+    fi
+    return 0
+}
+
+cmd_help() {
+    # Extract header comment block as help text
+    sed -n '2,/^$/p' "$0" | sed 's/^# \?//'
+    return 0
+}
+
+# Main dispatch
+main() {
+    local command="${1:-help}"
+    shift 2>/dev/null || true
+
+    case "$command" in
+        save)    cmd_save "$@" ;;
+        load)    cmd_load ;;
+        clear)   cmd_clear ;;
+        status)  cmd_status ;;
+        help|-h|--help) cmd_help ;;
+        *)
+            print_error "Unknown command: ${command}"
+            cmd_help
+            return 1
+            ;;
+    esac
+}
+
+main "$@"
diff --git a/.agents/scripts/supervisor-helper.sh b/.agents/scripts/supervisor-helper.sh
index 559baaf199..18643a7223 100755
--- a/.agents/scripts/supervisor-helper.sh
+++ b/.agents/scripts/supervisor-helper.sh
@@ -40,9 +40,9 @@
 #                                   -> failed     (dispatch failure / unrecoverable)
 #
 # Post-PR lifecycle (t128.8):
-#   complete -> pr_review -> merging -> merged -> deploying -> deployed
+#   complete -> pr_review -> review_triage -> merging -> merged -> deploying -> deployed
 #   Workers exit after PR creation. Supervisor handles remaining stages:
-#   wait for CI, merge (squash), postflight, deploy, worktree cleanup.
+#   wait for CI, triage review threads (t148), merge (squash), postflight, deploy, worktree cleanup.
 #
 # Outcome evaluation (3-tier):
 #   1. Deterministic: FULL_LOOP_COMPLETE/TASK_COMPLETE signals, EXIT codes
@@ -79,7 +79,7 @@ readonly MEMORY_HELPER="${SCRIPT_DIR}/memory-helper.sh"   # Used by pulse comman
 export MAIL_HELPER MEMORY_HELPER
 
 # Valid states for the state machine
-readonly VALID_STATES="queued dispatched running evaluating retrying complete pr_review merging merged deploying deployed blocked failed cancelled"
+readonly VALID_STATES="queued dispatched running evaluating retrying complete pr_review review_triage merging merged deploying deployed blocked failed cancelled"
 
 # Valid state transitions (from:to pairs)
 # Format: "from_state:to_state" - checked by validate_transition()
@@ -105,9 +105,14 @@ readonly -a VALID_TRANSITIONS=(
     # Post-PR lifecycle transitions (t128.8)
     "complete:pr_review"
     "complete:deployed"
+    "pr_review:review_triage"
     "pr_review:merging"
     "pr_review:blocked"
     "pr_review:cancelled"
+    # Review triage transitions (t148) - triage unresolved review threads before merge
+    "review_triage:merging"
+    "review_triage:blocked"
+    "review_triage:pr_review"
     "merging:merged"
     "merging:blocked"
     "merging:failed"
@@ -365,7 +370,7 @@ CREATE TABLE tasks (
     repo            TEXT NOT NULL,
     description     TEXT,
     status          TEXT NOT NULL DEFAULT 'queued'
-                    CHECK(status IN ('queued','dispatched','running','evaluating','retrying','complete','pr_review','merging','merged','deploying','deployed','blocked','failed','cancelled')),
+                    CHECK(status IN ('queued','dispatched','running','evaluating','retrying','complete','pr_review','review_triage','merging','merged','deploying','deployed','blocked','failed','cancelled')),
     session_id      TEXT,
     worktree        TEXT,
     branch          TEXT,
@@ -391,6 +396,44 @@ MIGRATE
         log_success "Database schema migrated for post-PR lifecycle states"
     fi
 
+    # Migrate: add review_triage to status CHECK constraint (t148)
+    if [[ -n "$check_sql" ]] && ! echo "$check_sql" | grep -q 'review_triage'; then
+        log_info "Migrating database schema for review_triage state (t148)..."
+        db "$SUPERVISOR_DB" << 'MIGRATE_TRIAGE'
+PRAGMA foreign_keys=OFF;
+BEGIN TRANSACTION;
+ALTER TABLE tasks RENAME TO tasks_old_t148;
+CREATE TABLE tasks (
+    id              TEXT PRIMARY KEY,
+    repo            TEXT NOT NULL,
+    description     TEXT,
+    status          TEXT NOT NULL DEFAULT 'queued'
+                    CHECK(status IN ('queued','dispatched','running','evaluating','retrying','complete','pr_review','review_triage','merging','merged','deploying','deployed','blocked','failed','cancelled')),
+    session_id      TEXT,
+    worktree        TEXT,
+    branch          TEXT,
+    log_file        TEXT,
+    retries         INTEGER NOT NULL DEFAULT 0,
+    max_retries     INTEGER NOT NULL DEFAULT 3,
+    model           TEXT DEFAULT 'anthropic/claude-opus-4-6',
+    error           TEXT,
+    pr_url          TEXT,
+    created_at      TEXT NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ','now')),
+    started_at      TEXT,
+    completed_at    TEXT,
+    updated_at      TEXT NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ','now'))
+);
+INSERT INTO tasks SELECT * FROM tasks_old_t148;
+DROP TABLE tasks_old_t148;
+CREATE INDEX IF NOT EXISTS idx_tasks_status ON tasks(status);
+CREATE INDEX IF NOT EXISTS idx_tasks_repo ON tasks(repo);
+CREATE INDEX IF NOT EXISTS idx_tasks_created ON tasks(created_at);
+COMMIT;
+PRAGMA foreign_keys=ON;
+MIGRATE_TRIAGE
+        log_success "Database schema migrated for review_triage state"
+    fi
+
     # Migrate: add max_load_factor column to batches if missing (t135.15.4)
     local has_max_load
     has_max_load=$(db "$SUPERVISOR_DB" "SELECT count(*) FROM pragma_table_info('batches') WHERE name='max_load_factor';" 2>/dev/null || echo "0")
@@ -426,7 +469,7 @@ CREATE TABLE IF NOT EXISTS tasks (
     repo            TEXT NOT NULL,
     description     TEXT,
     status          TEXT NOT NULL DEFAULT 'queued'
-                    CHECK(status IN ('queued','dispatched','running','evaluating','retrying','complete','pr_review','merging','merged','deploying','deployed','blocked','failed','cancelled')),
+                    CHECK(status IN ('queued','dispatched','running','evaluating','retrying','complete','pr_review','review_triage','merging','merged','deploying','deployed','blocked','failed','cancelled')),
     session_id      TEXT,
     worktree        TEXT,
     branch          TEXT,
@@ -2541,13 +2584,48 @@ cmd_reprompt() {
     if [[ -n "$prompt_override" ]]; then
         reprompt_msg="$prompt_override"
     else
-        reprompt_msg="The previous attempt for task $task_id encountered an issue: ${terror:-unknown error}.
+        # Detect if this was a compaction/context-limit exit (clean exit, no signal)
+        # and tailor the reprompt accordingly
+        local is_compaction_retry=false
+        if [[ "$terror" == *"clean_exit_no_signal"* || "$terror" == *"context"* ]]; then
+            is_compaction_retry=true
+        fi
+
+        if [[ "$is_compaction_retry" == "true" ]]; then
+            # Compaction-specific reprompt: the worker likely ran out of context
+            # window and exited cleanly without completing. Give it targeted guidance.
+            local worktree_status=""
+            if [[ -n "$tworktree" && -d "$tworktree" ]]; then
+                worktree_status=$(git -C "$tworktree" status --short 2>/dev/null | head -20 || echo "")
+                local recent_commits
+                recent_commits=$(git -C "$tworktree" log --oneline -5 2>/dev/null || echo "")
+                if [[ -n "$recent_commits" ]]; then
+                    worktree_status="${worktree_status}
+
+Recent commits on this branch:
+${recent_commits}"
+                fi
+            fi
+
+            reprompt_msg="The previous attempt for task $task_id likely hit context compaction and exited without completing.
+
+/full-loop $task_id
+
+IMPORTANT: Check the worktree for partial progress before starting fresh. The previous session may have made commits or uncommitted changes.
+${worktree_status:+
+Current worktree state:
+${worktree_status}}
+
+Task description: ${tdesc:-$task_id}"
+        else
+            reprompt_msg="The previous attempt for task $task_id encountered an issue: ${terror:-unknown error}.
 
 Please continue the /full-loop for $task_id. Pick up where the previous attempt left off.
 If the task was partially completed, verify what's done and continue from there.
 If it failed entirely, start fresh with /full-loop $task_id.
 
 Task description: ${tdesc:-$task_id}"
+        fi
     fi
 
     # Set up log file for this retry attempt
@@ -2778,6 +2856,157 @@ check_pr_status() {
     return 0
 }
 
+#######################################
+# Check for unresolved review threads on a PR (t148)
+# Returns JSON: {"total": N, "unresolved": N, "threads": [...]}
+# Each thread: {"author": "...", "body": "...", "is_bot": bool, "path": "...", "line": N}
+#######################################
+check_review_threads() {
+    local task_id="$1"
+
+    ensure_db
+
+    local escaped_id
+    escaped_id=$(sql_escape "$task_id")
+    local pr_url
+    pr_url=$(db "$SUPERVISOR_DB" "SELECT pr_url FROM tasks WHERE id = '$escaped_id';")
+
+    if [[ -z "$pr_url" || "$pr_url" == "no_pr" || "$pr_url" == "task_only" ]]; then
+        echo '{"total":0,"unresolved":0,"threads":[]}'
+        return 0
+    fi
+
+    local pr_number
+    pr_number=$(echo "$pr_url" | grep -oE '[0-9]+$' || echo "")
+    local repo_slug
+    repo_slug=$(echo "$pr_url" | grep -oE 'github\.com/[^/]+/[^/]+' | sed 's|github\.com/||' || echo "")
+
+    if [[ -z "$pr_number" || -z "$repo_slug" ]]; then
+        echo '{"total":0,"unresolved":0,"threads":[]}'
+        return 0
+    fi
+
+    local owner repo_name
+    owner="${repo_slug%%/*}"
+    repo_name="${repo_slug##*/}"
+
+    # Fetch review threads via GraphQL (REST API doesn't expose thread resolution)
+    local graphql_result
+    graphql_result=$(gh api graphql -f query='
+        query($owner: String!, $repo: String!, $pr: Int!) {
+            repository(owner: $owner, name: $repo) {
+                pullRequest(number: $pr) {
+                    reviewThreads(first: 100) {  # NOTE: max 100 per page; pagination not yet implemented (rare to exceed 100 threads)
+                        totalCount
+                        nodes {
+                            isResolved
+                            comments(first: 1) {
+                                nodes {
+                                    author { login }
+                                    body
+                                    path
+                                    line
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    ' -f owner="$owner" -f repo="$repo_name" -F pr="$pr_number" 2>/dev/null || echo "")
+
+    if [[ -z "$graphql_result" ]]; then
+        echo '{"total":0,"unresolved":0,"threads":[],"error":"graphql_failed"}'
+        return 0
+    fi
+
+    # Known bot accounts that leave COMMENTED reviews
+    local bot_pattern='coderabbitai|github-actions|dependabot|renovate|copilot|gemini|sonarcloud|codecov'
+
+    # Parse and classify threads
+    local result
+    result=$(echo "$graphql_result" | jq --arg bots "$bot_pattern" '
+        .data.repository.pullRequest.reviewThreads as $threads |
+        {
+            total: $threads.totalCount,
+            unresolved: [$threads.nodes[] | select(.isResolved == false)] | length,
+            threads: [
+                $threads.nodes[]
+                | select(.isResolved == false)
+                | .comments.nodes[0] as $c
+                | {
+                    author: ($c.author.login // "unknown"),
+                    body: ($c.body // "" | .[0:500]),
+                    is_bot: (($c.author.login // "") | test($bots; "i")),
+                    path: ($c.path // ""),
+                    line: ($c.line // 0)
+                }
+            ]
+        }
+    ' 2>/dev/null || echo '{"total":0,"unresolved":0,"threads":[]}')
+
+    echo "$result"
+    return 0
+}
+
+#######################################
+# Triage review feedback for a PR (t148)
+# Classifies unresolved threads by severity and source (bot vs human)
+# Returns: triage_pass (safe to merge), triage_block (needs attention), triage_skip (no threads)
+#######################################
+triage_review_feedback() {
+    local task_id="$1"
+    local skip_triage="${2:-false}"
+
+    if [[ "$skip_triage" == "true" ]]; then
+        echo "triage_skip"
+        return 0
+    fi
+
+    local threads_json
+    threads_json=$(check_review_threads "$task_id")
+
+    local unresolved
+    unresolved=$(echo "$threads_json" | jq -r '.unresolved // 0' 2>/dev/null || echo "0")
+
+    if [[ "$unresolved" -eq 0 ]]; then
+        log_info "Review triage for $task_id: no unresolved threads"
+        echo "triage_pass"
+        return 0
+    fi
+
+    # Count bot vs human threads
+    local bot_count human_count
+    bot_count=$(echo "$threads_json" | jq '[.threads[] | select(.is_bot == true)] | length' 2>/dev/null || echo "0")
+    human_count=$(echo "$threads_json" | jq '[.threads[] | select(.is_bot == false)] | length' 2>/dev/null || echo "0")
+
+    log_info "Review triage for $task_id: $unresolved unresolved ($bot_count bot, $human_count human)"
+
+    # Human review threads always block
+    if [[ "$human_count" -gt 0 ]]; then
+        log_warn "  $human_count human review thread(s) unresolved - blocking merge"
+        echo "$threads_json" | jq -r '.threads[] | select(.is_bot == false) | "  - \(.author) on \(.path):\(.line): \(.body[0:120])"' 2>/dev/null || true
+        echo "triage_block"
+        return 0
+    fi
+
+    # Bot-only threads: classify by severity keywords
+    local high_severity_count
+    high_severity_count=$(echo "$threads_json" | jq '[.threads[] | select(.is_bot == true) | select(.body | test("\\b(bug|security|vulnerability|critical|error|crash|data.loss|injection|XSS|CSRF)\\b"; "i"))] | length' 2>/dev/null || echo "0")
+
+    if [[ "$high_severity_count" -gt 0 ]]; then
+        log_warn "  $high_severity_count high-severity bot finding(s) - blocking merge"
+        echo "$threads_json" | jq -r '.threads[] | select(.is_bot == true) | select(.body | test("\\b(bug|security|vulnerability|critical|error|crash|data.loss|injection|XSS|CSRF)\\b"; "i")) | "  - \(.author) on \(.path):\(.line): \(.body[0:120])"' 2>/dev/null || true
+        echo "triage_block"
+        return 0
+    fi
+
+    # Bot-only, low severity (style, naming, docs suggestions) - pass with warning
+    log_info "  $bot_count bot thread(s) are low-severity - safe to merge"
+    echo "triage_pass"
+    return 0
+}
+
 #######################################
 # Command: pr-check - check PR CI/review status for a task
 #######################################
@@ -3033,7 +3262,7 @@ cleanup_after_merge() {
 # Checks CI, merges, runs postflight, deploys, cleans up worktree
 #######################################
 cmd_pr_lifecycle() {
-    local task_id="" dry_run="false"
+    local task_id="" dry_run="false" skip_review_triage="false"
 
     if [[ $# -gt 0 && ! "$1" =~ ^-- ]]; then
         task_id="$1"
@@ -3043,6 +3272,7 @@ cmd_pr_lifecycle() {
     while [[ $# -gt 0 ]]; do
         case "$1" in
             --dry-run) dry_run=true; shift ;;
+            --skip-review-triage) skip_review_triage=true; shift ;;
             *) log_error "Unknown option: $1"; return 1 ;;
         esac
     done
@@ -3115,10 +3345,38 @@ cmd_pr_lifecycle() {
 
         case "$pr_status" in
             ready_to_merge)
-                if [[ "$dry_run" == "false" ]]; then
-                    cmd_transition "$task_id" "merging" 2>/dev/null || true
+                # Review triage gate (t148): check unresolved threads before merge
+                if [[ "$skip_review_triage" == "false" && "$dry_run" == "false" ]]; then
+                    cmd_transition "$task_id" "review_triage" 2>/dev/null || true
+                    local triage_result
+                    triage_result=$(triage_review_feedback "$task_id")
+                    case "$triage_result" in
+                        triage_pass|triage_skip)
+                            log_info "Review triage passed for $task_id"
+                            cmd_transition "$task_id" "merging" 2>/dev/null || true
+                            tstatus="merging"
+                            ;;
+                        triage_block)
+                            log_warn "Review triage blocked $task_id - unresolved review threads need attention"
+                            cmd_transition "$task_id" "blocked" --error "Unresolved review threads (t148 triage)" 2>/dev/null || true
+                            send_task_notification "$task_id" "blocked" "Unresolved review threads need attention before merge" 2>/dev/null || true
+                            return 1
+                            ;;
+                        *)
+                            log_warn "Review triage returned unexpected result: $triage_result"
+                            cmd_transition "$task_id" "merging" 2>/dev/null || true
+                            tstatus="merging"
+                            ;;
+                    esac
+                else
+                    if [[ "$skip_review_triage" == "true" ]]; then
+                        log_warn "Review triage bypassed via --skip-review-triage for $task_id"
+                    fi
+                    if [[ "$dry_run" == "false" ]]; then
+                        cmd_transition "$task_id" "merging" 2>/dev/null || true
+                    fi
+                    tstatus="merging"
                 fi
-                tstatus="merging"
                 ;;
             already_merged)
                 if [[ "$dry_run" == "false" ]]; then
@@ -3160,10 +3418,11 @@ cmd_pr_lifecycle() {
                 ;;
             no_pr)
                 # Track consecutive no_pr failures to avoid infinite retry loop
-                local no_pr_key="no_pr_retries_${task_id}"
                 local no_pr_count
-                no_pr_count=$(db "SELECT COALESCE(
-                    (SELECT CAST(json_extract(error, '$.no_pr_retries') AS INTEGER)
+                no_pr_count=$(db "$SUPERVISOR_DB" "SELECT COALESCE(
+                    (SELECT CAST(json_extract(
+                        CASE WHEN json_valid(error) THEN error ELSE '{}' END,
+                        '$.no_pr_retries') AS INTEGER)
                      FROM tasks WHERE id='$task_id'), 0);" 2>/dev/null || echo "0")
                 no_pr_count=$((no_pr_count + 1))
 
@@ -3180,12 +3439,45 @@ cmd_pr_lifecycle() {
 
                 log_warn "No PR found for $task_id (attempt $no_pr_count/5)"
                 # Store retry count in error field as JSON
-                db "UPDATE tasks SET error = json_set(COALESCE(error, '{}'), '$.no_pr_retries', $no_pr_count), updated_at = strftime('%Y-%m-%dT%H:%M:%SZ','now') WHERE id='$task_id';" 2>/dev/null || true
+                db "$SUPERVISOR_DB" "UPDATE tasks SET error = json_set(
+                    CASE WHEN json_valid(error) THEN error ELSE '{}' END,
+                    '$.no_pr_retries', $no_pr_count),
+                    updated_at = strftime('%Y-%m-%dT%H:%M:%SZ','now')
+                    WHERE id='$task_id';" 2>/dev/null || true
                 return 0
                 ;;
         esac
     fi
 
+    # Step 2b: Re-triage if already in review_triage state (t148)
+    if [[ "$tstatus" == "review_triage" ]]; then
+        local triage_result
+        triage_result=$(triage_review_feedback "$task_id" "$skip_review_triage")
+        case "$triage_result" in
+            triage_pass|triage_skip)
+                log_info "Review triage passed for $task_id"
+                if [[ "$dry_run" == "false" ]]; then
+                    cmd_transition "$task_id" "merging" 2>/dev/null || true
+                fi
+                tstatus="merging"
+                ;;
+            triage_block)
+                log_warn "Review triage still blocking $task_id"
+                if [[ "$dry_run" == "false" ]]; then
+                    cmd_transition "$task_id" "blocked" --error "Unresolved review threads (t148 triage)" 2>/dev/null || true
+                fi
+                return 1
+                ;;
+            *)
+                log_warn "Review triage returned unexpected: $triage_result - proceeding"
+                if [[ "$dry_run" == "false" ]]; then
+                    cmd_transition "$task_id" "merging" 2>/dev/null || true
+                fi
+                tstatus="merging"
+                ;;
+        esac
+    fi
+
     # Step 3: Merge
     if [[ "$tstatus" == "merging" ]]; then
         if [[ "$dry_run" == "true" ]]; then
@@ -3245,7 +3537,7 @@ process_post_pr_lifecycle() {
     ensure_db
 
     # Find tasks eligible for post-PR processing
-    local where_clause="t.status IN ('complete', 'pr_review', 'merging', 'merged', 'deploying')"
+    local where_clause="t.status IN ('complete', 'pr_review', 'review_triage', 'merging', 'merged', 'deploying')"
     if [[ -n "$batch_id" ]]; then
         where_clause="$where_clause AND EXISTS (SELECT 1 FROM batch_tasks bt WHERE bt.task_id = t.id AND bt.batch_id = '$(sql_escape "$batch_id")')"
     fi
@@ -3549,7 +3841,7 @@ cmd_pulse() {
     local total_complete
     total_complete=$(db "$SUPERVISOR_DB" "SELECT count(*) FROM tasks WHERE status IN ('complete', 'deployed');")
     local total_pr_review
-    total_pr_review=$(db "$SUPERVISOR_DB" "SELECT count(*) FROM tasks WHERE status IN ('pr_review', 'merging', 'merged', 'deploying');")
+    total_pr_review=$(db "$SUPERVISOR_DB" "SELECT count(*) FROM tasks WHERE status IN ('pr_review', 'review_triage', 'merging', 'merged', 'deploying');")
 
     local total_failed
     total_failed=$(db "$SUPERVISOR_DB" "SELECT count(*) FROM tasks WHERE status IN ('failed', 'blocked');")
@@ -3632,6 +3924,14 @@ cmd_pulse() {
         log_info "  Cleaned:    $orphan_killed stale worker processes"
     fi
 
+    # Phase 5: Persist pulse checkpoint for orchestrator compaction resilience
+    # Writes state to disk so an AI orchestrating session can re-orient after
+    # context compaction by reading this file. Cron pulses are stateless bash
+    # and don't need this, but interactive orchestrators do.
+    save_pulse_checkpoint "${batch_id:-}" "$total_running" "$total_queued" \
+        "$total_complete" "$total_tasks" "$total_failed" "$total_pr_review" \
+        "$completed_count" "$dispatched_count" 2>/dev/null || true
+
     return 0
 }
 
@@ -3824,6 +4124,80 @@ update_todo_on_complete() {
     }
 
     log_success "Committed and pushed TODO.md update for $task_id"
+
+    # Sync matching GitHub issue (close on completion)
+    sync_github_issue "$task_id" "complete" 2>/dev/null || true
+
+    return 0
+}
+
+#######################################
+# Sync GitHub issue state for a task
+# Closes matching issue on completion, adds blocked comment on failure
+# Convention: issue titles prefixed with "t{NNN}: "
+# TODO.md tasks reference issues with "ref:GH#{NNN}"
+#######################################
+sync_github_issue() {
+    local task_id="$1"
+    local event_type="$2"  # complete, blocked
+    local detail="${3:-}"
+
+    # Require gh CLI
+    if ! command -v gh &>/dev/null; then
+        log_info "gh CLI not available, skipping GitHub issue sync"
+        return 0
+    fi
+
+    ensure_db
+
+    local escaped_id
+    escaped_id=$(sql_escape "$task_id")
+    local trepo
+    trepo=$(db "$SUPERVISOR_DB" "SELECT repo FROM tasks WHERE id = '$escaped_id';" 2>/dev/null || echo "")
+
+    if [[ -z "$trepo" ]]; then
+        return 0
+    fi
+
+    # Determine repo slug from git remote
+    local repo_slug
+    repo_slug=$(git -C "$trepo" remote get-url origin 2>/dev/null | grep -oE '[^/:]+/[^/.]+' | tail -1 || echo "")
+    if [[ -z "$repo_slug" ]]; then
+        return 0
+    fi
+
+    # Search for matching issue by t-number prefix in title
+    local issue_number
+    issue_number=$(gh issue list --repo "$repo_slug" --state open --search "${task_id}:" --json number,title --jq ".[] | select(.title | startswith(\"${task_id}:\") or startswith(\"${task_id} \")) | .number" 2>/dev/null | head -1 || echo "")
+
+    if [[ -z "$issue_number" ]]; then
+        log_info "No matching GitHub issue found for $task_id"
+        return 0
+    fi
+
+    case "$event_type" in
+        complete)
+            local pr_url
+            pr_url=$(db "$SUPERVISOR_DB" "SELECT pr_url FROM tasks WHERE id = '$escaped_id';" 2>/dev/null || echo "")
+            local close_comment="Completed by supervisor."
+            if [[ -n "$pr_url" && "$pr_url" != "no_pr" && "$pr_url" != "task_only" ]]; then
+                close_comment="Completed via ${pr_url}"
+            fi
+            gh issue close "$issue_number" --repo "$repo_slug" --comment "$close_comment" 2>/dev/null || {
+                log_warn "Failed to close GitHub issue #$issue_number for $task_id"
+                return 1
+            }
+            log_success "Closed GitHub issue #$issue_number for $task_id"
+            ;;
+        blocked)
+            gh issue comment "$issue_number" --repo "$repo_slug" --body "Blocked by supervisor: ${detail}" 2>/dev/null || {
+                log_warn "Failed to comment on GitHub issue #$issue_number for $task_id"
+                return 1
+            }
+            log_info "Added blocked comment to GitHub issue #$issue_number for $task_id"
+            ;;
+    esac
+
     return 0
 }
 
@@ -3917,6 +4291,10 @@ ${notes_line}" "$todo_file"
     }
 
     log_success "Committed and pushed TODO.md blocked update for $task_id"
+
+    # Sync matching GitHub issue (add blocked comment)
+    sync_github_issue "$task_id" "blocked" "$reason" 2>/dev/null || true
+
     return 0
 }
 
@@ -3999,6 +4377,134 @@ send_task_notification() {
     return 0
 }
 
+#######################################
+# Persist pulse state to disk for orchestrator compaction resilience
+# An AI session orchestrating batch tasks can read this file after context
+# compaction to re-orient without losing track of progress.
+# File: $SUPERVISOR_DIR/pulse-checkpoint.md
+#######################################
+save_pulse_checkpoint() {
+    local batch_id="$1"
+    local running="$2"
+    local queued="$3"
+    local complete="$4"
+    local total="$5"
+    local failed="$6"
+    local pr_review="$7"
+    local just_completed="$8"
+    local just_dispatched="$9"
+
+    local checkpoint_file="$SUPERVISOR_DIR/pulse-checkpoint.md"
+    local timestamp
+    timestamp="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+
+    # Collect per-task state for running/queued tasks
+    local running_details
+    running_details=$(db -separator '|' "$SUPERVISOR_DB" "
+        SELECT id, description, worktree, status
+        FROM tasks WHERE status IN ('running', 'dispatched', 'evaluating')
+        ORDER BY started_at ASC;
+    " 2>/dev/null || echo "")
+
+    local queued_details
+    queued_details=$(db -separator '|' "$SUPERVISOR_DB" "
+        SELECT id, description
+        FROM tasks WHERE status = 'queued'
+        ORDER BY rowid ASC LIMIT 10;
+    " 2>/dev/null || echo "")
+
+    local blocked_details
+    blocked_details=$(db -separator '|' "$SUPERVISOR_DB" "
+        SELECT id, description, error
+        FROM tasks WHERE status IN ('blocked', 'failed')
+        ORDER BY updated_at DESC LIMIT 5;
+    " 2>/dev/null || echo "")
+
+    local pr_details
+    pr_details=$(db -separator '|' "$SUPERVISOR_DB" "
+        SELECT id, description, status, pr_url
+        FROM tasks WHERE status IN ('pr_review', 'review_triage', 'merging', 'merged', 'deploying')
+        ORDER BY updated_at DESC;
+    " 2>/dev/null || echo "")
+
+    # Build checkpoint
+    {
+        echo "# Supervisor Pulse Checkpoint"
+        echo ""
+        echo "Updated: ${timestamp}"
+        echo "Batch: ${batch_id:-all}"
+        echo ""
+        echo "## Summary"
+        echo ""
+        echo "| Metric | Count |"
+        echo "|--------|-------|"
+        echo "| Running | ${running} |"
+        echo "| Queued | ${queued} |"
+        echo "| Complete | ${complete} |"
+        echo "| Failed/Blocked | ${failed} |"
+        echo "| Post-PR | ${pr_review} |"
+        echo "| Total | ${total} |"
+        echo "| Just Completed | ${just_completed} |"
+        echo "| Just Dispatched | ${just_dispatched} |"
+        echo ""
+
+        if [[ -n "$running_details" ]]; then
+            echo "## Running Tasks"
+            echo ""
+            while IFS='|' read -r tid tdesc tworktree tstatus; do
+                echo "- **${tid}** (${tstatus}): ${tdesc}"
+                if [[ -n "$tworktree" ]]; then
+                    echo "  Worktree: ${tworktree}"
+                fi
+            done <<< "$running_details"
+            echo ""
+        fi
+
+        if [[ -n "$queued_details" ]]; then
+            echo "## Next Queued (up to 10)"
+            echo ""
+            while IFS='|' read -r tid tdesc; do
+                echo "- **${tid}**: ${tdesc}"
+            done <<< "$queued_details"
+            echo ""
+        fi
+
+        if [[ -n "$pr_details" ]]; then
+            echo "## Post-PR Lifecycle"
+            echo ""
+            while IFS='|' read -r tid tdesc tstatus tpr; do
+                echo "- **${tid}** (${tstatus}): ${tdesc}"
+                if [[ -n "$tpr" ]]; then
+                    echo "  PR: ${tpr}"
+                fi
+            done <<< "$pr_details"
+            echo ""
+        fi
+
+        if [[ -n "$blocked_details" ]]; then
+            echo "## Blocked/Failed (recent)"
+            echo ""
+            while IFS='|' read -r tid tdesc terror; do
+                echo "- **${tid}**: ${tdesc}"
+                if [[ -n "$terror" ]]; then
+                    echo "  Error: ${terror}"
+                fi
+            done <<< "$blocked_details"
+            echo ""
+        fi
+
+        echo "## Re-orientation Instructions"
+        echo ""
+        echo "If you are an AI orchestrator reading this after context compaction:"
+        echo "1. Run \`supervisor-helper.sh status\` for full current state"
+        echo "2. Run \`supervisor-helper.sh pulse\` to trigger the next cycle"
+        echo "3. Check \`supervisor-helper.sh cron status\` to verify cron is running"
+        echo "4. Blocked/failed tasks may need manual intervention"
+    } > "$checkpoint_file"
+
+    return 0
+}
+
 #######################################
 # Send a macOS notification for batch progress milestones
 # Called from pulse summary when notable progress occurs
@@ -4768,10 +5274,12 @@ State Machine (worker lifecycle):
                                   -> blocked    (needs human input / max retries exceeded)
                                   -> failed     (dispatch failure / unrecoverable)
 
-Post-PR Lifecycle (t128.8 - supervisor handles directly, no worker needed):
-  complete -> pr_review -> merging -> merged -> deploying -> deployed
+Post-PR Lifecycle (t128.8 + t148 review triage):
+  complete -> pr_review -> review_triage -> merging -> merged -> deploying -> deployed
   Workers exit after PR creation. Supervisor detects complete tasks with PR URLs
-  and handles: CI wait, merge (squash), postflight, deploy, worktree cleanup.
+  and handles: CI wait, review thread triage, merge (squash), postflight, deploy,
+  worktree cleanup. Review triage checks unresolved threads via GraphQL and blocks
+  merge if human reviews or high-severity bot findings are unaddressed.
 
 Outcome Evaluation (3-tier):
   1. Deterministic: FULL_LOOP_COMPLETE/TASK_COMPLETE signals, EXIT codes
@@ -4816,6 +5324,10 @@ Options for 'evaluate':
 Options for 'pulse':
   --batch <batch_id>     Only pulse tasks in this batch
 
+Options for 'pr-lifecycle':
+  --dry-run              Show what would happen without doing it
+  --skip-review-triage   Skip review thread triage (merge without checking threads)
+
 Options for 'cleanup':
   --dry-run              Show what would be cleaned without doing it
 
diff --git a/.agents/scripts/x-helper.sh b/.agents/scripts/x-helper.sh
new file mode 100644
index 0000000000..16476f16f2
--- /dev/null
+++ b/.agents/scripts/x-helper.sh
@@ -0,0 +1,189 @@
+#!/usr/bin/env bash
+# x-helper.sh - Fetch X/Twitter posts via fxtwitter API (no auth required)
+# Part of aidevops framework: https://aidevops.sh
+#
+# Usage:
+#   x-helper.sh [command] [options]
+#
+# Commands:
+#   fetch <url>       Fetch a tweet/post by URL (x.com or twitter.com)
+#   thread <url>      Fetch a thread starting from the given post
+#   user <handle>     Fetch recent posts from a user (limited)
+#   help              Show this help
+#
+# Options:
+#   --json            Output raw JSON
+#   --format md       Output as markdown (default)
+#   --format text     Output as plain text
+#
+# Uses fxtwitter.com API (no authentication required).
+# Ref: https://github.com/FixTweet/FxTwitter
+
+set -euo pipefail
+
+readonly FXTWITTER_API="https://api.fxtwitter.com"
+
+readonly GREEN='\033[0;32m'
+readonly YELLOW='\033[1;33m'
+readonly RED='\033[0;31m'
+readonly NC='\033[0m'
+
+print_success() { printf "${GREEN}%s${NC}\n" "$1"; }
+print_warning() { printf "${YELLOW}%s${NC}\n" "$1"; }
+print_error() { printf "${RED}%s${NC}\n" "$1" >&2; }
+
+extract_tweet_path() {
+    local url="$1"
+    # Extract username/status/id from various URL formats
+    # Handles: x.com/user/status/123, twitter.com/user/status/123
+    local path
+    path=$(echo "$url" | sed -E 's|https?://(x\.com|twitter\.com)/||' | sed 's|[?#].*||')
+    echo "$path"
+}
+
+extract_username() {
+    local url="$1"
+    echo "$url" | sed -E 's|https?://(x\.com|twitter\.com)/||' | sed 's|/.*||' | sed 's|^@||'
+}
+
+cmd_fetch() {
+    local url="${1:-}"
+    local output_format="${2:-md}"
+    local raw_json="${3:-false}"
+
+    if [[ -z "$url" ]]; then
+        print_error "URL required. Usage: x-helper.sh fetch <url>"
+        return 1
+    fi
+
+    local path
+    path=$(extract_tweet_path "$url")
+
+    local response
+    response=$(curl -fsS --max-time 20 --retry 2 --retry-connrefused \
+        "${FXTWITTER_API}/${path}" 2>/dev/null) || {
+        print_error "Failed to fetch tweet"
+        return 1
+    }
+
+    if [[ "$raw_json" == "true" ]]; then
+        echo "$response" | jq . 2>/dev/null || echo "$response"
+        return 0
+    fi
+
+    # Parse JSON and format output
+    local author author_handle text created_at likes retweets replies
+    author=$(echo "$response" | jq -r '.tweet.author.name // "Unknown"' 2>/dev/null || echo "Unknown")
+    author_handle=$(echo "$response" | jq -r '.tweet.author.screen_name // "unknown"' 2>/dev/null || echo "unknown")
+    text=$(echo "$response" | jq -r '.tweet.text // ""' 2>/dev/null || echo "")
+    created_at=$(echo "$response" | jq -r '.tweet.created_at // ""' 2>/dev/null || echo "")
+    likes=$(echo "$response" | jq -r '.tweet.likes // 0' 2>/dev/null || echo "0")
+    retweets=$(echo "$response" | jq -r '.tweet.retweets // 0' 2>/dev/null || echo "0")
+    replies=$(echo "$response" | jq -r '.tweet.replies // 0' 2>/dev/null || echo "0")
+
+    if [[ "$output_format" == "text" ]]; then
+        echo "@${author_handle} (${author})"
+        echo "${created_at}"
+        echo ""
+        echo "${text}"
+        echo ""
+        echo "Likes: ${likes} | Retweets: ${retweets} | Replies: ${replies}"
+    else
+        # Markdown format
+        printf "**%s** (@%s)\n" "$author" "$author_handle"
+        printf "*%s*\n\n" "$created_at"
+        printf "%s\n\n" "$text"
+        printf "Likes: %s | Retweets: %s | Replies: %s\n" "$likes" "$retweets" "$replies"
+        printf "\nSource: %s\n" "$url"
+    fi
+
+    return 0
+}
+
+cmd_thread() {
+    local url="${1:-}"
+
+    if [[ -z "$url" ]]; then
+        print_error "URL required. Usage: x-helper.sh thread <url>"
+        return 1
+    fi
+
+    # fxtwitter doesn't have a native thread endpoint, but we can fetch
+    # the conversation by following reply chains
+    print_warning "Thread fetching is limited - fxtwitter returns individual posts"
+    print_warning "Fetching initial post..."
+    echo ""
+    cmd_fetch "$url" "md" "false"
+
+    return 0
+}
+
+cmd_user() {
+    local handle="${1:-}"
+
+    if [[ -z "$handle" ]]; then
+        print_error "Handle required. Usage: x-helper.sh user <handle>"
+        return 1
+    fi
+
+    # Remove @ prefix if present
+    handle="${handle#@}"
+
+    local response
+    response=$(curl -fsS --max-time 20 --retry 2 --retry-connrefused \
+        "${FXTWITTER_API}/${handle}" 2>/dev/null) || {
+        print_error "Failed to fetch user profile"
+        return 1
+    }
+
+    # Parse user info
+    local name followers following description
+    name=$(echo "$response" | jq -r '.user.name // "Unknown"' 2>/dev/null || echo "Unknown")
+    followers=$(echo "$response" | jq -r '.user.followers // 0' 2>/dev/null || echo "0")
+    following=$(echo "$response" | jq -r '.user.following // 0' 2>/dev/null || echo "0")
+    description=$(echo "$response" | jq -r '.user.description // ""' 2>/dev/null || echo "")
+
+    printf "**%s** (@%s)\n" "$name" "$handle"
+    printf "Followers: %s | Following: %s\n\n" "$followers" "$following"
+    if [[ -n "$description" ]]; then
+        printf "%s\n" "$description"
+    fi
+
+    return 0
+}
+
+cmd_help() {
+    sed -n '2,/^$/p' "$0" | sed 's/^# \?//'
+    return 0
+}
+
+main() {
+    local command="${1:-help}"
+    shift 2>/dev/null || true
+
+    local output_format="md"
+    local raw_json="false"
+    local args=()
+
+    while [[ $# -gt 0 ]]; do
+        case "$1" in
+            --json) raw_json="true"; shift ;;
+            --format) [[ $# -lt 2 ]] && { print_error "--format requires a value"; return 1; }; output_format="$2"; shift 2 ;;
+            *) args+=("$1"); shift ;;
+        esac
+    done
+
+    case "$command" in
+        fetch)   cmd_fetch "${args[0]:-}" "$output_format" "$raw_json" ;;
+        thread)  cmd_thread "${args[0]:-}" ;;
+        user)    cmd_user "${args[0]:-}" ;;
+        help|-h|--help) cmd_help ;;
+        *)
+            print_error "Unknown command: ${command}"
+            cmd_help
+            return 1
+            ;;
+    esac
+}
+
+main "$@"
diff --git a/.agents/seo/backlink-checker.md b/.agents/seo/backlink-checker.md
new file mode 100644
index 0000000000..08dcbee156
--- /dev/null
+++ b/.agents/seo/backlink-checker.md
@@ -0,0 +1,105 @@
+---
+description: Backlink monitoring and expired domain discovery for link reclamation
+mode: subagent
+tools:
+  read: true
+  write: true
+  edit: true
+  bash: true
+  glob: true
+  grep: true
+  webfetch: true
+  task: true
+---
+
+# Backlink & Expired Domain Checker
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Monitor backlinks, detect lost/broken links, find expired referring domains for purchase
+- **Data Sources**: Ahrefs API, DataForSEO Backlinks API, WHOIS lookups
+- **Helpers**: `scripts/seo-export-ahrefs.sh`, `scripts/seo-export-dataforseo.sh` (backlink data export)
+
+**Workflow**: Fetch backlink profile -> Identify lost/broken links -> Check domain expiry status -> Rank by DA/DR/traffic value -> Output purchase candidates
+
+<!-- AI-CONTEXT-END -->
+
+## Data Sources
+
+### Ahrefs API (Primary)
+
+> See `scripts/seo-export-ahrefs.sh` for the export implementation.
+
+Ahrefs endpoints used:
+- `/v3/site-explorer/all-backlinks` - Full backlink list
+- `/v3/site-explorer/backlinks-new-lost` - New/lost link changes
+- `/v3/site-explorer/referring-domains` - Unique referring domains
+
+### DataForSEO Backlinks API (Alternative)
+
+> See `scripts/seo-export-dataforseo.sh` for the export implementation.
+
+DataForSEO endpoints:
+- `/v3/backlinks/backlinks/live` - Live backlink data
+- `/v3/backlinks/referring_domains/live` - Referring domains
+- `/v3/backlinks/bulk_new_lost_backlinks/live` - Bulk new/lost
+
+## Expired Domain Detection
+
+### WHOIS Lookup
+
+```bash
+# Check if a referring domain has expired
+whois example-referrer.com | grep -i "expir"
+
+# Batch check (pipe from backlink export)
+seo-helper.sh backlinks example.com --referring-domains-only | while read -r domain; do
+    expiry=$(whois "$domain" 2>/dev/null | grep -i "expiry\|expiration" | head -1)
+    echo "$domain: $expiry"
+done
+```
+
+### Domain Availability Tools
+
+| Tool | Type | Notes |
+|------|------|-------|
+| `whois` CLI | Free | Rate-limited, varies by TLD |
+| expired-domains.co | Web | Aggregates expired domain lists |
+| expireddomains.net | Web | Filters by DA, backlinks, age |
+| GoDaddy Auctions API | API | Auction/aftermarket domains |
+| Namecheap API | API | Registration availability check |
+
+### GitHub Tools for Expired Domain Discovery
+
+- [peterprototypes/expired-domains](https://github.com/peterprototypes/expired-domains) - Rust CLI for checking domain expiry
+- [Jeongseup/expired-domain-finder](https://github.com/Jeongseup/expired-domain-finder) - Python bulk checker
+
+## Reclamation Workflow
+
+1. **Export referring domains** with DR/DA scores
+2. **Filter lost/broken** links from last 90 days
+3. **WHOIS check** each lost referring domain
+4. **Score candidates** by:
+   - Domain Rating (DR) or Domain Authority (DA)
+   - Number of backlinks the domain had
+   - Traffic estimate (Ahrefs/SimilarWeb)
+   - Registration cost vs. link value
+5. **Output ranked list** of purchase candidates
+
+The full reclamation pipeline combines backlink export with WHOIS checks. Use the Ahrefs or DataForSEO export scripts to get lost backlinks, then pipe through WHOIS lookups above.
+
+## Integration with aidevops
+
+- Pairs with `seo/ahrefs.md` for API access
+- Pairs with `seo/dataforseo.md` for alternative data source
+- Pairs with `seo/domain-research.md` for DNS intelligence on candidates
+- Results can feed into `seo/link-building.md` workflows
+
+## Related
+
+- `seo/ahrefs.md` - Ahrefs API integration
+- `seo/dataforseo.md` - DataForSEO API integration
+- `seo/domain-research.md` - DNS reconnaissance
+- `seo/link-building.md` - Link-building strategies
diff --git a/.agents/subagent-index.toon b/.agents/subagent-index.toon
index ae76630ac9..3b26dbb473 100644
--- a/.agents/subagent-index.toon
+++ b/.agents/subagent-index.toon
@@ -99,6 +99,7 @@ todo-ready.sh,Show tasks with no open blockers
 ralph-loop-helper.sh,Iterative AI development loops
 full-loop-helper.sh,End-to-end development loop (task to PR to deploy)
 session-review-helper.sh,Gather session context for review
+session-checkpoint-helper.sh,Persist session state to survive context compaction
 mail-helper.sh,TOON-based inter-agent mailbox system
 memory-helper.sh,Cross-session memory (SQLite FTS5)
 memory-embeddings-helper.sh,Semantic memory search with vector embeddings (opt-in)
diff --git a/.agents/tools/accounts/subscription-audit.md b/.agents/tools/accounts/subscription-audit.md
new file mode 100644
index 0000000000..787f7bd061
--- /dev/null
+++ b/.agents/tools/accounts/subscription-audit.md
@@ -0,0 +1,147 @@
+---
+description: Subscription audit - discover, track, and optimize recurring payments
+mode: subagent
+tools:
+  read: true
+  write: true
+  edit: true
+  bash: true
+  glob: true
+  grep: true
+  webfetch: true
+  task: true
+---
+
+# Subscription Audit
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Discover all active subscriptions, calculate total spend, identify savings
+- **Command**: `/subscription-audit` or `@subscription-audit`
+- **Data Sources**: Email receipts, bank statements (CSV), manual inventory
+
+**Quick start**: Use the `/subscription-audit` command to start an interactive audit. The agent guides you through discovery, analysis, and optimization phases.
+
+<!-- AI-CONTEXT-END -->
+
+## Audit Workflow
+
+### Phase 1: Discovery
+
+Identify all recurring charges from multiple sources:
+
+#### Email Receipt Scanning
+
+Search email for common subscription receipt patterns:
+
+```bash
+# Gmail search queries (via IMAP or Gmail API)
+# Recurring payment receipts
+"receipt" OR "invoice" OR "subscription" OR "renewal" OR "billing"
+
+# Common SaaS senders
+from:(noreply@github.com OR billing@stripe.com OR receipts@paddle.com)
+
+# Subscription keywords
+"your subscription" OR "monthly charge" OR "annual renewal" OR "auto-renewal"
+```
+
+#### Bank Statement Import
+
+Import CSV bank statements (auto-detects format for Chase, Amex, Barclays, Monzo, Revolut, generic CSV). The agent identifies recurring patterns (same merchant, similar amount, regular interval).
+
+#### Manual Inventory
+
+Common categories to check:
+
+| Category | Examples |
+|----------|----------|
+| **Dev Tools** | GitHub, GitLab, JetBrains, Vercel, Netlify, Railway, Fly.io |
+| **Cloud/Infra** | AWS, GCP, Azure, DigitalOcean, Hetzner, Cloudflare |
+| **AI/ML** | OpenAI, Anthropic, Google AI, Replicate, HuggingFace |
+| **Domains** | Namecheap, Cloudflare, GoDaddy, Porkbun |
+| **Email** | Google Workspace, Fastmail, Proton, Mailgun, SendGrid |
+| **Monitoring** | Datadog, Sentry, PagerDuty, UptimeRobot, BetterStack |
+| **Security** | 1Password, Bitwarden, NordVPN, Tailscale |
+| **Productivity** | Notion, Linear, Slack, Zoom, Figma, Miro |
+| **Media** | Spotify, YouTube Premium, Netflix, Audible |
+| **Storage** | iCloud, Dropbox, Backblaze, Wasabi |
+
+### Phase 2: Analysis
+
+The analysis phase generates a report that includes:
+- **Monthly total**: Sum of all recurring charges
+- **Annual projection**: Monthly * 12 + annual-only subscriptions
+- **Category breakdown**: Spend per category
+- **Unused detection**: Subscriptions with no recent login/API activity
+- **Duplicate detection**: Multiple tools serving same purpose
+- **Price increase alerts**: Charges that increased vs. last period
+
+### Phase 3: Optimization
+
+| Strategy | Savings Potential |
+|----------|-------------------|
+| Cancel unused subscriptions | 10-30% |
+| Downgrade overprovisioned tiers | 5-15% |
+| Switch to annual billing | 15-20% per service |
+| Consolidate overlapping tools | 10-25% |
+| Use open-source alternatives | Variable |
+| Negotiate enterprise discounts | 10-40% |
+
+### Open-Source Alternatives
+
+| Paid Tool | Free Alternative |
+|-----------|------------------|
+| GitHub Copilot | Claude Code (free tier), Cody |
+| Notion | Obsidian, Logseq |
+| Slack | Mattermost, Zulip |
+| Datadog | Grafana + Prometheus |
+| PagerDuty | Grafana OnCall |
+| 1Password | Bitwarden (self-hosted) |
+| Vercel | Coolify (self-hosted) |
+| Linear | Plane (self-hosted) |
+
+## Report Format
+
+```text
+Subscription Audit Report
+=========================
+Generated: 2026-02-07
+Period: Monthly
+
+Active Subscriptions: 23
+Monthly Total: $847.50
+Annual Projection: $10,170.00
+
+By Category:
+  Cloud/Infra:    $312.00  (36.8%)
+  Dev Tools:      $189.00  (22.3%)
+  AI/ML:          $145.00  (17.1%)
+  Productivity:    $89.50  (10.6%)
+  Security:        $52.00   (6.1%)
+  Other:           $60.00   (7.1%)
+
+Recommendations:
+1. [UNUSED] Datadog Pro - no dashboards viewed in 60 days ($99/mo)
+2. [DUPLICATE] Both Sentry and BetterStack for error tracking ($38/mo overlap)
+3. [DOWNGRADE] GitHub Team → Free (only 2 private repos) ($4/mo)
+4. [ANNUAL] Switch Vercel to annual billing (save $36/yr)
+
+Potential Monthly Savings: $141.00 (16.6%)
+```
+
+## Data Storage
+
+Subscription data stored in SQLite at `~/.aidevops/.agent-workspace/work/subscriptions/subscriptions.db`.
+
+Schema:
+- `subscriptions` - Active subscriptions (name, category, amount, interval, provider)
+- `audit_history` - Past audit snapshots for trend tracking
+- `recommendations` - Generated recommendations with status (applied/dismissed)
+
+## Related
+
+- `accounts.md` - Financial operations agent
+- `tools/credentials/api-key-setup.md` - API key management (related spend)
diff --git a/.agents/tools/context/rapidfuzz.md b/.agents/tools/context/rapidfuzz.md
new file mode 100644
index 0000000000..ac482ced8f
--- /dev/null
+++ b/.agents/tools/context/rapidfuzz.md
@@ -0,0 +1,121 @@
+---
+description: RapidFuzz - fast fuzzy string matching library for Python and C++
+mode: subagent
+tools:
+  read: true
+  write: false
+  edit: false
+  bash: true
+  glob: true
+  grep: true
+---
+
+# RapidFuzz - Fuzzy String Matching
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Fast fuzzy string matching (Levenshtein, Jaro-Winkler, token-based)
+- **Install**: `pip install rapidfuzz` or `conda install -c conda-forge rapidfuzz`
+- **Repo**: https://github.com/rapidfuzz/RapidFuzz (2.8k+ stars, MIT)
+- **Docs**: https://rapidfuzz.github.io/RapidFuzz/
+
+**When to use**: Deduplication, typo-tolerant search, record linkage, autocomplete, entity matching. 10-100x faster than `fuzzywuzzy` (C++ backend, no Python-only fallback).
+
+<!-- AI-CONTEXT-END -->
+
+## Core Functions
+
+### Simple Ratio (character-level similarity)
+
+```python
+from rapidfuzz import fuzz
+
+fuzz.ratio("fuzzy wuzzy", "fuzzy wuzzy")       # 100.0
+fuzz.ratio("fuzzy wuzzy", "fzzy wzzy")         # ~84.2
+```
+
+### Partial Ratio (substring matching)
+
+```python
+fuzz.partial_ratio("fuzzy wuzzy", "wuzzy")     # 100.0
+```
+
+### Token Sort Ratio (order-independent)
+
+```python
+fuzz.token_sort_ratio("New York Mets", "Mets New York")  # 100.0
+```
+
+### Token Set Ratio (handles duplicates and extras)
+
+```python
+fuzz.token_set_ratio(
+    "fuzzy wuzzy was a bear",
+    "fuzzy fuzzy was a bear"
+)  # High score despite differences
+```
+
+## Process Module (batch matching)
+
+```python
+from rapidfuzz import process
+
+choices = ["Atlanta Falcons", "New York Jets", "New York Giants", "Dallas Cowboys"]
+
+# Best match
+process.extractOne("new york jets", choices)
+# ('New York Jets', 100.0, 1)
+
+# Top N matches
+process.extract("new york jets", choices, limit=2)
+# [('New York Jets', 100.0, 1), ('New York Giants', 78.57, 2)]
+
+# Score cutoff
+process.extract("new york jets", choices, score_cutoff=80)
+```
+
+## Distance Functions
+
+```python
+from rapidfuzz.distance import Levenshtein, Jaro, JaroWinkler
+
+Levenshtein.distance("kitten", "sitting")       # 3
+Levenshtein.normalized_similarity("kitten", "sitting")  # ~0.571
+
+Jaro.similarity("martha", "marhta")             # ~0.944
+JaroWinkler.similarity("martha", "marhta")      # ~0.961
+```
+
+## Performance Tips
+
+- Use `score_cutoff` parameter to skip low-scoring comparisons early
+- Use `process.cdist()` for pairwise distance matrices (NumPy integration)
+- Pre-process strings with `rapidfuzz.utils.default_process` (lowercase, strip)
+- For large datasets (>10k items), use `process.extract` with `workers=-1` for parallelism
+
+## Common Patterns in aidevops
+
+```python
+# Fuzzy-match user input to known commands
+from rapidfuzz import process
+commands = ["deploy", "status", "update", "rollback", "logs"]
+match, score, idx = process.extractOne(user_input, commands)
+if score > 80:
+    execute(match)
+
+# Deduplicate similar entries
+from rapidfuzz import fuzz
+def deduplicate(items, threshold=85):
+    unique = []
+    for item in items:
+        if not any(fuzz.ratio(item, u) > threshold for u in unique):
+            unique.append(item)
+    return unique
+```
+
+## Related
+
+- `tools/context/mcp-discovery.md` - Uses fuzzy matching for tool search
+- `memory-helper.sh` - Could use RapidFuzz for similar memory deduplication
diff --git a/.agents/tools/document/document-extraction.md b/.agents/tools/document/document-extraction.md
new file mode 100644
index 0000000000..6a69dd8397
--- /dev/null
+++ b/.agents/tools/document/document-extraction.md
@@ -0,0 +1,200 @@
+---
+description: Privacy-preserving document extraction with Docling, ExtractThinker, and Presidio
+mode: subagent
+tools:
+  read: true
+  write: true
+  edit: true
+  bash: true
+  glob: true
+  grep: true
+  webfetch: true
+  task: true
+---
+
+# Document Extraction
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Extract structured data from documents (PDF, DOCX, images) with PII redaction
+- **Stack**: Docling (parsing) + ExtractThinker (LLM extraction) + Presidio (PII detection)
+- **Privacy**: Fully local processing via Ollama or Cloudflare Workers AI
+- **PRD**: `todo/tasks/prd-document-extraction.md`
+
+**Status**: Planned (see `todo/tasks/prd-document-extraction.md` for PRD). The Python components below are available for direct use; a CLI helper script is not yet implemented.
+
+<!-- AI-CONTEXT-END -->
+
+## Architecture
+
+```text
+Document Input (PDF/DOCX/Image/HTML)
+         │
+    ┌────┴────┐
+    │ Docling  │  ← Document parsing (layout, tables, OCR)
+    └────┬────┘
+         │
+    ┌────┴──────────┐
+    │ ExtractThinker │  ← LLM-powered structured extraction
+    └────┬──────────┘
+         │
+    ┌────┴────────┐
+    │  Presidio    │  ← PII detection and redaction (optional)
+    └────┬────────┘
+         │
+    Structured Output (JSON/CSV/Markdown)
+```
+
+## Components
+
+### Docling (Document Parsing)
+
+IBM's document conversion library. Handles complex layouts, tables, and OCR.
+
+```bash
+pip install docling
+
+# Python usage
+from docling.document_converter import DocumentConverter
+converter = DocumentConverter()
+result = converter.convert("document.pdf")
+print(result.document.export_to_markdown())
+```
+
+- **Formats**: PDF, DOCX, PPTX, XLSX, HTML, images, AsciiDoc
+- **Features**: Table extraction, OCR (EasyOCR/Tesseract), layout analysis
+- **Repo**: https://github.com/DS4SD/docling
+
+### ExtractThinker (LLM Extraction)
+
+Pydantic-based structured extraction using LLMs.
+
+```python
+from extract_thinker import Extractor, Contract
+from pydantic import BaseModel
+
+class Invoice(BaseModel):
+    vendor: str
+    date: str
+    total: float
+    items: list[dict]
+
+extractor = Extractor()
+extractor.load_document_loader("docling")
+extractor.load_llm("ollama/llama3.2")  # Local model
+
+result = extractor.extract("invoice.pdf", Invoice)
+print(result.model_dump_json(indent=2))
+```
+
+- **Repo**: https://github.com/enoch3712/ExtractThinker
+- **LLM backends**: Ollama (local), OpenAI, Anthropic, Google, Cloudflare Workers AI
+
+### Presidio (PII Redaction)
+
+Microsoft's PII detection and anonymization.
+
+```python
+from presidio_analyzer import AnalyzerEngine
+from presidio_anonymizer import AnonymizerEngine
+
+analyzer = AnalyzerEngine()
+anonymizer = AnonymizerEngine()
+
+text = "John Smith's SSN is 123-45-6789"
+results = analyzer.analyze(text=text, language="en")
+anonymized = anonymizer.anonymize(text=text, analyzer_results=results)
+print(anonymized.text)  # "<PERSON>'s SSN is <US_SSN>"
+```
+
+- **Entities**: PERSON, EMAIL, PHONE, SSN, CREDIT_CARD, IBAN, IP_ADDRESS, etc.
+- **Repo**: https://github.com/microsoft/presidio
+
+## Extraction Schemas (Templates)
+
+> These are example/template schemas for common document types. Customize for your project.
+
+### Invoice
+
+```python
+class Invoice(BaseModel):
+    vendor_name: str
+    vendor_address: str | None
+    invoice_number: str
+    invoice_date: str
+    due_date: str | None
+    subtotal: float
+    tax: float | None
+    total: float
+    currency: str
+    line_items: list[LineItem]
+```
+
+### Receipt
+
+```python
+class Receipt(BaseModel):
+    merchant: str
+    date: str
+    total: float
+    payment_method: str | None
+    items: list[ReceiptItem]
+```
+
+### Contract
+
+```python
+class ContractSummary(BaseModel):
+    parties: list[str]
+    effective_date: str
+    termination_date: str | None
+    key_terms: list[str]
+    obligations: list[str]
+```
+
+## Privacy Modes
+
+| Mode | LLM | PII Handling | Use Case |
+|------|-----|-------------|----------|
+| **Local** | Ollama (llama3.2) | Presidio redact before LLM | Maximum privacy |
+| **Edge** | Cloudflare Workers AI | Presidio redact before API | Good privacy, faster |
+| **Cloud** | OpenAI/Anthropic | Presidio redact before API | Best quality |
+| **None** | Any | No redaction | Non-sensitive documents |
+
+## Installation
+
+```bash
+# Core
+pip install docling extract-thinker
+
+# PII (optional)
+pip install presidio-analyzer presidio-anonymizer
+python -m spacy download en_core_web_lg
+
+# Local LLM (optional)
+brew install ollama && ollama pull llama3.2
+
+# OCR backends (optional)
+pip install easyocr  # or: brew install tesseract
+```
+
+## When to Use (vs Unstract)
+
+| Feature | This Stack (Docling+ExtractThinker) | Unstract MCP |
+|---------|-------------------------------------|-------------|
+| **Privacy** | Full local processing via Ollama | Cloud or self-hosted |
+| **Schema control** | Pydantic models, custom schemas | Pre-built extractors |
+| **PII redaction** | Built-in via Presidio | Manual |
+| **Setup** | pip install, no server | Docker/server required |
+| **Best for** | Custom extraction pipelines | Quick document processing |
+
+Use this stack when you need custom schemas, PII redaction, or fully local processing. Use Unstract when you need quick setup with pre-built extractors.
+
+## Related
+
+- `tools/document/pandoc-helper.sh` - Document format conversion
+- `tools/document/mineru.md` - Alternative PDF extraction (MinerU)
+- `services/document-processing/unstract.md` - Self-hosted document processing (alternative)
+- `todo/tasks/prd-document-extraction.md` - Full PRD
diff --git a/.agents/tools/terminal/terminal-optimization.md b/.agents/tools/terminal/terminal-optimization.md
new file mode 100644
index 0000000000..f99f107436
--- /dev/null
+++ b/.agents/tools/terminal/terminal-optimization.md
@@ -0,0 +1,182 @@
+---
+description: Terminal environment optimization - shell config, aliases, performance tuning
+mode: subagent
+tools:
+  read: true
+  write: true
+  edit: true
+  bash: true
+  glob: true
+  grep: true
+---
+
+# Terminal Optimization
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Audit and optimize terminal environment (shell startup, aliases, PATH, tools)
+- **Command**: `/terminal-optimize` or `@terminal-optimization`
+- **Scope**: Shell config (.bashrc/.zshrc), PATH dedup, tool recommendations, startup profiling
+
+**Quick audit**:
+
+```bash
+# Profile shell startup time
+time zsh -i -c exit    # or: time bash -i -c exit
+
+# Count PATH entries (duplicates waste lookup time)
+echo "$PATH" | tr ':' '\n' | sort | uniq -c | sort -rn | head
+
+# Check shell config size
+wc -l ~/.zshrc ~/.bashrc ~/.profile 2>/dev/null
+```
+
+<!-- AI-CONTEXT-END -->
+
+## Optimization Checklist
+
+### 1. Shell Startup Time
+
+Target: <200ms for interactive shell.
+
+```bash
+# Profile zsh startup
+zsh -xvs 2>&1 | ts -i '%.s' | head -50
+
+# Profile bash startup
+bash --norc --noprofile -c 'time bash -i -c exit'
+
+# Common slow culprits:
+# - nvm (Node Version Manager) - lazy-load instead
+# - conda init - lazy-load instead
+# - rbenv/pyenv init - lazy-load instead
+# - oh-my-zsh plugins - reduce to essentials
+# - compinit (zsh completion) - cache with zcompdump
+```
+
+### Lazy-Loading Pattern
+
+```bash
+# Instead of: eval "$(nvm init)"
+# Use lazy-load:
+nvm() {
+    unset -f nvm node npm npx
+    export NVM_DIR="$HOME/.nvm"
+    [ -s "$NVM_DIR/nvm.sh" ] && source "$NVM_DIR/nvm.sh"
+    nvm "$@"
+}
+```
+
+### 2. PATH Optimization
+
+```bash
+# Deduplicate PATH
+typeset -U PATH  # zsh built-in
+# or for bash:
+PATH=$(echo "$PATH" | tr ':' '\n' | awk '!seen[$0]++' | tr '\n' ':' | sed 's/:$//')
+
+# Remove non-existent directories
+PATH=$(echo "$PATH" | tr ':' '\n' | while read -r dir; do
+    [ -d "$dir" ] && echo "$dir"
+done | tr '\n' ':' | sed 's/:$//')
+```
+
+### 3. Modern Tool Replacements
+
+| Classic | Modern | Speedup | Install |
+|---------|--------|---------|---------|
+| `find` | `fd` | 5-10x | `brew install fd` |
+| `grep` | `ripgrep (rg)` | 5-20x | `brew install ripgrep` |
+| `cat` | `bat` | Syntax highlighting | `brew install bat` |
+| `ls` | `eza` | Better output | `brew install eza` |
+| `du` | `dust` | Visual tree | `brew install dust` |
+| `top` | `btop` | Better UI | `brew install btop` |
+| `diff` | `delta` | Syntax-aware | `brew install git-delta` |
+| `sed` | `sd` | Simpler syntax | `brew install sd` |
+| `curl` | `xh` | Colorized HTTP | `brew install xh` |
+| `man` | `tldr` | Quick examples | `brew install tldr` |
+
+### 4. Shell Aliases
+
+```bash
+# Git shortcuts
+alias gs='git status'
+alias gd='git diff'
+alias gl='git log --oneline -20'
+alias gp='git pull --rebase'
+
+# Navigation
+alias ..='cd ..'
+alias ...='cd ../..'
+alias ll='eza -la --git'
+
+# Safety
+alias rm='rm -i'
+alias cp='cp -i'
+alias mv='mv -i'
+
+# aidevops
+alias ad='aidevops'
+alias ads='aidevops status'
+alias adu='aidevops update'
+```
+
+### 5. Completion and History
+
+```bash
+# Zsh: better history search
+bindkey '^R' history-incremental-search-backward
+HISTSIZE=50000
+SAVEHIST=50000
+setopt SHARE_HISTORY
+setopt HIST_IGNORE_DUPS
+setopt HIST_IGNORE_SPACE
+
+# fzf integration (fuzzy history search)
+# brew install fzf && $(brew --prefix)/opt/fzf/install
+```
+
+### 6. Terminal Multiplexer
+
+```bash
+# tmux essentials
+brew install tmux
+
+# Minimal .tmux.conf
+set -g mouse on
+set -g history-limit 50000
+set -g default-terminal "screen-256color"
+bind | split-window -h -c "#{pane_current_path}"
+bind - split-window -v -c "#{pane_current_path}"
+```
+
+## Audit Report Format
+
+When running `/terminal-optimize`, output:
+
+```text
+Terminal Optimization Report
+============================
+Shell: zsh 5.9
+Startup time: 342ms (target: <200ms)
+PATH entries: 23 (4 duplicates, 2 missing)
+Config size: .zshrc 187 lines
+
+Issues Found:
+1. [SLOW] nvm init adds 180ms - suggest lazy-load
+2. [DUP] /usr/local/bin appears 3x in PATH
+3. [MISSING] /opt/homebrew/bin not in PATH but Homebrew installed
+4. [TOOL] Using grep instead of ripgrep (5-20x slower)
+
+Recommendations:
+1. Lazy-load nvm (saves ~180ms)
+2. Deduplicate PATH (saves ~5ms lookup)
+3. Install: fd, ripgrep, bat, eza
+4. Add fzf for fuzzy history search
+```
+
+## Related
+
+- `tools/terminal/terminal-title.md` - Terminal title management
diff --git a/.agents/tools/voice/transcription.md b/.agents/tools/voice/transcription.md
new file mode 100644
index 0000000000..50f19284ca
--- /dev/null
+++ b/.agents/tools/voice/transcription.md
@@ -0,0 +1,204 @@
+---
+description: Audio/video transcription with local and cloud models
+mode: subagent
+tools:
+  read: true
+  write: true
+  edit: true
+  bash: true
+  glob: true
+  grep: true
+  webfetch: true
+  task: true
+---
+
+# Audio/Video Transcription
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: Transcribe audio/video from YouTube, URLs, or local files
+- **Helper**: `transcription-helper.sh [transcribe|models|configure] [options]`
+- **Default model**: Whisper Large v3 Turbo (best speed/accuracy tradeoff)
+- **Dependencies**: `yt-dlp` (YouTube), `ffmpeg` (audio extraction), `faster-whisper` or `whisper.cpp` (local)
+
+**Quick Commands**:
+
+```bash
+# Transcribe YouTube video
+transcription-helper.sh transcribe "https://youtu.be/dQw4w9WgXcQ"
+
+# Transcribe local file
+transcription-helper.sh transcribe recording.mp3
+
+# Transcribe with specific model
+transcription-helper.sh transcribe recording.mp3 --model large-v3-turbo
+
+# List available models
+transcription-helper.sh models
+```
+
+<!-- AI-CONTEXT-END -->
+
+## Input Sources
+
+| Source | Detection | Extraction |
+|--------|-----------|------------|
+| YouTube URL | `youtu.be/` or `youtube.com/watch` | `yt-dlp -x --audio-format wav` |
+| Direct media URL | HTTP(S) with media extension | `curl` + `ffmpeg` if video |
+| Local audio | `.wav`, `.mp3`, `.flac`, `.ogg`, `.m4a` | Direct input |
+| Local video | `.mp4`, `.mkv`, `.webm`, `.avi` | `ffmpeg -i input -vn -acodec pcm_s16le` |
+
+## Local Models (Whisper Family)
+
+### via faster-whisper (Recommended)
+
+CTranslate2-based, 4x faster than OpenAI Whisper with same accuracy.
+See `voice-bridge.py:99-115` for the repo's `FasterWhisperSTT` implementation.
+
+```bash
+pip install faster-whisper
+```
+
+Official usage: https://github.com/SYSTRAN/faster-whisper#usage
+
+### via whisper.cpp (C++ native)
+
+Optimized for Apple Silicon and CPU inference.
+
+```bash
+# Build from source (no Homebrew formula)
+git clone https://github.com/ggml-org/whisper.cpp.git
+cd whisper.cpp && cmake -B build && cmake --build build -j --config Release
+
+# Download model
+sh ./models/download-ggml-model.sh large-v3-turbo
+
+# Transcribe
+./build/bin/whisper-cli -f audio.wav -otxt -osrt
+```
+
+### Model Comparison
+
+| Model | Size | Speed | Accuracy | Notes |
+|-------|------|-------|----------|-------|
+| Tiny | 75MB | 9.5 | 6.0-6.5 | Draft/preview only |
+| Base | 142MB | 8.5 | 7.2-7.5 | Quick transcription |
+| Small | 461MB | 7.0 | 8.5 | Good balance |
+| Medium | 1.5GB | 5.0 | 9.0 | Solid quality |
+| Large v2 | 2.9GB | 3.0 | 9.6 | High quality |
+| Large v3 | 2.9GB | 3.0 | 9.8 | Best quality |
+| **Large v3 Turbo** | **1.5GB** | **7.5** | **9.7** | **Recommended default** |
+| Large v3 Turbo Q | 547MB | 7.5 | 9.5 | Quantized, smaller |
+
+### Other Local Models
+
+| Model | Size | Speed | Accuracy | Notes |
+|-------|------|-------|----------|-------|
+| NVIDIA Parakeet V2 | 474MB | 9.9 | 9.4 | English-only, fastest |
+| NVIDIA Parakeet V3 | 494MB | 9.9 | 9.4 | Multilingual, experimental |
+| Apple Speech | Built-in | 9.0 | 9.0 | macOS 26+, on-device |
+
+## Cloud APIs
+
+| Provider | Model | Accuracy | Speed | Cost |
+|----------|-------|----------|-------|------|
+| **Groq** | Whisper Large v3 Turbo | 9.6 | Lightning | Free tier available |
+| **ElevenLabs** | Scribe v2 | 9.9 | Fast | Pay per minute |
+| **ElevenLabs** | Scribe v1 | 9.8 | Fast | Pay per minute |
+| **Mistral** | Voxtral Mini | 9.7 | Fast | Pay per token |
+| **Deepgram** | Nova-2 | 9.5 | Real-time | Pay per minute |
+| **Deepgram** | Nova-3 Medical | 9.6 | Real-time | English-only, clinical |
+| **OpenAI** | Whisper API | 9.5 | Fast | $0.006/min |
+| **Google** | Gemini 3 Pro | 9.7 | Fast | Multimodal input |
+| **Google** | Gemini 3 Flash | 9.5 | Fastest | Low latency |
+| **Soniox** | stt-async-v3 | 9.6 | Async | Batch processing |
+
+### Groq (Fastest Cloud)
+
+```bash
+curl https://api.groq.com/openai/v1/audio/transcriptions \
+  -H "Authorization: Bearer ${GROQ_API_KEY}" \
+  -H "Content-Type: multipart/form-data" \
+  -F "file=@audio.wav" \
+  -F "model=whisper-large-v3" \
+  -F "response_format=verbose_json"
+```
+
+Docs: https://console.groq.com/docs/speech-text
+
+### OpenAI Whisper API
+
+```bash
+curl https://api.openai.com/v1/audio/transcriptions \
+  -H "Authorization: Bearer ${OPENAI_API_KEY}" \
+  -F "file=@audio.wav" \
+  -F "model=whisper-1" \
+  -F "response_format=srt"
+```
+
+Docs: https://platform.openai.com/docs/api-reference/audio/createTranscription
+
+### ElevenLabs Scribe
+
+```bash
+curl -X POST "https://api.elevenlabs.io/v1/speech-to-text" \
+  -H "xi-api-key: ${ELEVENLABS_API_KEY}" \
+  -F "file=@audio.wav" \
+  -F "model_id=scribe_v1"
+```
+
+Docs: https://elevenlabs.io/docs/api-reference/speech-to-text
+
+## Output Formats
+
+| Format | Extension | Use Case |
+|--------|-----------|----------|
+| Plain text | `.txt` | Reading, search indexing |
+| SRT | `.srt` | Video subtitles |
+| VTT | `.vtt` | Web video subtitles |
+| JSON | `.json` | Programmatic access, timestamps |
+| TSV | `.tsv` | Spreadsheet analysis |
+
+## Transcription Pipeline
+
+```text
+Input Source
+    │
+    ├── YouTube URL ──→ yt-dlp -x ──→ audio.wav
+    ├── Video URL ────→ curl + ffmpeg ──→ audio.wav
+    ├── Video file ───→ ffmpeg -vn ──→ audio.wav
+    └── Audio file ───→ (direct) ──→ audio.wav
+                                        │
+                                   Model Selection
+                                        │
+                              ┌─────────┴─────────┐
+                              │                    │
+                         Local Model          Cloud API
+                     (faster-whisper)      (Groq/OpenAI/etc)
+                              │                    │
+                              └─────────┬──────────┘
+                                        │
+                                   Output Format
+                                   (txt/srt/vtt/json)
+```
+
+## Dependencies
+
+```bash
+# Core
+brew install yt-dlp ffmpeg    # macOS
+apt install yt-dlp ffmpeg     # Ubuntu/Debian
+
+# Local inference (pick one)
+pip install faster-whisper     # Python (recommended)
+# whisper.cpp: build from source (see above)
+```
+
+## Related
+
+- `tools/voice/voice-models.md` - TTS models for speech generation
+- `tools/voice/speech-to-speech.md` - Full voice pipeline
+- `tools/content/summarize.md` - Can summarize transcribed content
+- `voice-helper.sh` - CLI for voice operations
diff --git a/.agents/tools/voice/voice-models.md b/.agents/tools/voice/voice-models.md
new file mode 100644
index 0000000000..7ab980055c
--- /dev/null
+++ b/.agents/tools/voice/voice-models.md
@@ -0,0 +1,103 @@
+---
+description: Voice AI models for speech generation (TTS) and transcription (STT)
+mode: subagent
+tools:
+  read: true
+  write: false
+  edit: false
+  bash: true
+  glob: true
+  grep: true
+  webfetch: true
+  task: true
+---
+
+# Voice AI Models
+
+<!-- AI-CONTEXT-START -->
+
+## Quick Reference
+
+- **Purpose**: TTS (text-to-speech) and STT (speech-to-text) model selection and usage
+- **Local TTS**: EdgeTTS (default), macOS Say, FacebookMMS — see `voice-bridge.py:133-238`
+- **Local STT**: See `tools/voice/transcription.md` for full transcription guide
+- **Cloud APIs**: ElevenLabs, OpenAI TTS — see official API docs
+
+**When to use**: Voice interfaces, content narration, accessibility, voice cloning, podcast generation, phone bots (with Twilio via `speech-to-speech.md`).
+
+<!-- AI-CONTEXT-END -->
+
+## Text-to-Speech (TTS) Models
+
+### Implemented in This Repo
+
+The voice bridge (`voice-bridge.py`) implements three TTS engines:
+
+#### EdgeTTS (Default)
+
+Microsoft Edge TTS via `edge-tts` package. See `voice-bridge.py:133-179`.
+
+- Free, no API key needed
+- 300+ voices across 70+ languages
+- Streaming support, adjustable rate
+- Default voice: `en-GB-SoniaNeural`
+
+```bash
+# Use via voice bridge
+voice-helper.sh talk
+```
+
+#### macOS Say
+
+Native macOS speech synthesis. See `voice-bridge.py:182-205`.
+
+- Built-in, zero dependencies
+- Default voice: `Samantha`
+- macOS only
+
+#### FacebookMMS TTS
+
+Meta's Massively Multilingual Speech. See `voice-bridge.py:208-238`.
+
+- 1,100+ languages
+- Requires `transformers` package
+- CPU-friendly
+
+### Other Local Models (Not Implemented)
+
+These are available but not integrated into the voice bridge:
+
+| Model | Notes | Docs |
+|-------|-------|------|
+| Qwen3-TTS | Multilingual, voice cloning | https://github.com/QwenLM/Qwen3-TTS |
+| Piper TTS | Lightweight, CPU-friendly, 100+ voices | https://github.com/rhasspy/piper |
+| Bark (Suno) | Expressive, non-speech sounds | https://github.com/suno-ai/bark |
+| Coqui TTS | Multi-model toolkit, voice cloning | https://github.com/coqui-ai/TTS |
+
+### Cloud APIs (Not Implemented)
+
+These require API keys and are not integrated into the voice bridge:
+
+| Provider | Docs |
+|----------|------|
+| ElevenLabs | https://elevenlabs.io/docs/api-reference/text-to-speech |
+| OpenAI TTS | https://platform.openai.com/docs/api-reference/audio/createSpeech |
+| Hugging Face Inference | https://huggingface.co/docs/api-inference/tasks/text-to-speech |
+
+## Model Selection Guide
+
+| Priority | Implemented | Not Yet Integrated |
+|----------|-------------|-------------------|
+| Default (free) | EdgeTTS | — |
+| macOS native | macOS Say | — |
+| Multilingual | FacebookMMS | Qwen3-TTS |
+| Voice clone | — | Qwen3-TTS, ElevenLabs |
+| Expressiveness | — | Bark |
+| CPU-only | All three | Piper |
+| Highest quality | — | ElevenLabs |
+
+## Related
+
+- `tools/voice/speech-to-speech.md` - Full voice pipeline (VAD+STT+LLM+TTS)
+- `tools/voice/transcription.md` - STT/transcription models
+- `voice-helper.sh` - CLI for voice operations
diff --git a/.agents/workflows/plans.md b/.agents/workflows/plans.md
index fe7f76cb76..ff70421c15 100644
--- a/.agents/workflows/plans.md
+++ b/.agents/workflows/plans.md
@@ -730,6 +730,52 @@ Uncommitted TODO changes are invisible to other sessions, agents, and the `/read
 | Status update | `chore: update task t{NNN} status` |
 | Plan creation | `chore: add plan for {title}` |
 
+## GitHub Issue Sync
+
+Tasks and GitHub issues MUST stay in sync with matching identifiers.
+
+### Convention
+
+- **GitHub issue titles** MUST be prefixed with their TODO.md task ID: `t{NNN}: {title}`
+- **TODO.md tasks** MUST reference their GitHub issue: `ref:GH#{NNN}`
+- Both directions must be maintained whenever either is created or updated
+
+### When Creating a GitHub Issue
+
+```bash
+# Always prefix with t-number
+gh issue create --title "t146: bug: supervisor no_pr retry counter non-functional" ...
+```
+
+### When Creating a TODO.md Task from an Issue
+
+```markdown
+- [ ] t146 bug: supervisor no_pr retry counter #bugfix ~15m logged:2026-02-07 ref:GH#439
+```
+
+### When Creating Both Together
+
+1. Assign the next available t-number
+2. Create the GitHub issue with `t{NNN}: ` prefix in the title
+3. Add the task to TODO.md with `ref:GH#{issue_number}`
+4. Commit and push TODO.md immediately
+
+### Automated Enforcement
+
+The supervisor's `update_todo_on_complete()` and `send_task_notification()` functions should maintain this sync. When the supervisor creates issues or updates TODO.md, it must:
+
+1. Check if a matching GitHub issue exists (search by `t{NNN}` in title)
+2. If not, create one with the `t{NNN}: ` prefix
+3. If the TODO.md task lacks `ref:GH#`, add it after issue creation
+4. When closing a task, close the matching GitHub issue with a comment
+
+### Why This Matters
+
+Without consistent t-number prefixes on issues, there's no way to:
+- Quickly find the GitHub issue for a TODO.md task (or vice versa)
+- Automate status sync between the two systems
+- Track which issues correspond to which batch of work
+
 ## Integration with Other Workflows
 
 | Workflow | Integration |
diff --git a/.agents/workflows/session-manager.md b/.agents/workflows/session-manager.md
index ecaf1dd766..f7f077c6e6 100644
--- a/.agents/workflows/session-manager.md
+++ b/.agents/workflows/session-manager.md
@@ -243,6 +243,56 @@ Suggestions:
 ---
 ```
 
+## Compaction Resilience (Long Autonomous Sessions)
+
+During long autonomous sessions (1h+), context compaction can cause loss of task state. Use the checkpoint system to persist state to disk.
+
+### Checkpoint Workflow
+
+```bash
+# After completing each task, save checkpoint:
+session-checkpoint-helper.sh save \
+  --task "t135.9" \
+  --next "t135.11,t014,t025" \
+  --worktree "/path/to/worktree" \
+  --batch "batch2-quality" \
+  --note "Completed trap cleanup for 29 scripts" \
+  --elapsed "90" \
+  --target "240"
+
+# Before starting any new task (especially after compaction), reload:
+session-checkpoint-helper.sh load
+
+# Check if checkpoint is stale:
+session-checkpoint-helper.sh status
+```
+
+### When to Checkpoint
+
+| Event | Action |
+|-------|--------|
+| Task completed | `save` with updated --task and --next |
+| PR created/merged | `save` with --note describing PR state |
+| Batch of files committed | `save` with --note listing what changed |
+| Before large operation | `save` as recovery point |
+| After context compaction | `load` to re-orient |
+
+### Self-Prompting Loop Pattern
+
+For autonomous multi-hour sessions, follow this loop after each task:
+
+1. Mark task complete in TODO.md
+2. Save checkpoint to disk
+3. Re-read checkpoint file (forces re-orientation after compaction)
+4. Read TODO.md for next task
+5. Start next task
+
+This ensures the agent always has current state even if context was compacted between steps 2 and 3.
+
+### Continuation Prompt Generation
+
+If a session must be resumed manually (new conversation), the checkpoint file at `~/.aidevops/.agent-workspace/tmp/session-checkpoint.md` contains enough state to generate a continuation prompt. The user can paste it or the agent can read it on startup.
+
 ## Related
 
 **AGENTS.md is the single source of truth for agent behavior.** This document is supplementary and defers to AGENTS.md where they differ.