prizmkit 1.1.91 → 1.1.93

package/bundled/dev-pipeline/run-feature.sh

@@ -9,14 +9,14 @@ set -euo pipefail
 # from a feature list.
 #
 # Usage:
-#   ./run-feature.sh run [.prizmkit/plans/feature-list.json]            Run all features
+#   ./run-feature.sh run [.prizmkit/plans/feature-list.json] [options]  Run all features
 #   ./run-feature.sh run <feature-id> [options]         Run a single feature
 #   ./run-feature.sh status [.prizmkit/plans/feature-list.json]         Show pipeline status
 #   ./run-feature.sh reset                              Clear all state
 #
 # Environment Variables:
-#   MAX_RETRIES           Max retries per feature (default: 3)
-#   SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)
+#   MAX_RETRIES           Max code retries per feature (default: 3)
+#   MAX_INFRA_RETRIES     Max infrastructure/provider retries per feature (default: 3)
 #   AI_CLI                AI CLI command name (override; also readable from .prizmkit/config.json)
 #   CODEBUDDY_CLI         Legacy alias for AI_CLI (deprecated, use AI_CLI instead)
 #   PRIZMKIT_PLATFORM     Platform hint for custom AI_CLI wrappers: codebuddy, claude, or codex
@@ -25,9 +25,6 @@ set -euo pipefail
 #   HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)
 #   STALE_KILL_THRESHOLD   Auto-kill session after N seconds of no progress (default: 900)
 #   HEARTBEAT_STALE_THRESHOLD  Heartbeat stale threshold in seconds (default: 600)
-#   LOG_CLEANUP_ENABLED   Run periodic log cleanup (default: 1)
-#   LOG_RETENTION_DAYS    Delete logs older than N days (default: 14)
-#   LOG_MAX_TOTAL_MB      Keep total logs under N MB via oldest-first cleanup (default: 1024)
 #   PIPELINE_MODE         Override mode for all features: lite|standard|full (used by daemon)
 #   DEV_BRANCH            Custom dev branch name (default: auto-generated dev/{feature_id}-YYYYMMDDHHmm)
 #   AUTO_PUSH             Auto-push to remote after successful feature (default: 0). Set to 1 to enable.
@@ -45,13 +42,10 @@ SCRIPTS_DIR="$SCRIPT_DIR/scripts"
 
 # Configuration (override via environment variables)
 MAX_RETRIES=${MAX_RETRIES:-3}
-SESSION_TIMEOUT=${SESSION_TIMEOUT:-0}
+MAX_INFRA_RETRIES=${MAX_INFRA_RETRIES:-3}
 HEARTBEAT_STALE_THRESHOLD=${HEARTBEAT_STALE_THRESHOLD:-600}
 HEARTBEAT_INTERVAL=${HEARTBEAT_INTERVAL:-30}
 STALE_KILL_THRESHOLD=${STALE_KILL_THRESHOLD:-900}
-LOG_CLEANUP_ENABLED=${LOG_CLEANUP_ENABLED:-1}
-LOG_RETENTION_DAYS=${LOG_RETENTION_DAYS:-14}
-LOG_MAX_TOTAL_MB=${LOG_MAX_TOTAL_MB:-1024}
 VERBOSE=${VERBOSE:-0}
 MODEL=${MODEL:-""}
 PRIZMKIT_EFFORT=${PRIZMKIT_EFFORT:-""}
@@ -124,13 +118,6 @@ spawn_and_wait_session() {
     start_progress_parser "$session_log" "$progress_json" "$SCRIPTS_DIR"
     local parser_pid="${_PARSER_PID:-}"
 
-    # Timeout watchdog (only if SESSION_TIMEOUT > 0)
-    local watcher_pid=""
-    if [[ $SESSION_TIMEOUT -gt 0 ]]; then
-        ( sleep "$SESSION_TIMEOUT" && kill -TERM "$cbc_pid" 2>/dev/null ) &
-        watcher_pid=$!
-    fi
-
     # Heartbeat monitor (reads progress.json when available, falls back to tail)
     # Also monitors for stale sessions and auto-kills if no progress for STALE_KILL_THRESHOLD seconds
     start_heartbeat "$cbc_pid" "$session_log" "$progress_json" "$HEARTBEAT_INTERVAL" "$STALE_KILL_THRESHOLD"
@@ -144,27 +131,35 @@ spawn_and_wait_session() {
         exit_code=$?
     fi
 
-    # Clean up watcher, heartbeat, and parser
-    [[ -n "$watcher_pid" ]] && kill "$watcher_pid" 2>/dev/null || true
+    # Clean up heartbeat and parser
     stop_heartbeat "$heartbeat_pid"
     stop_progress_parser "$parser_pid"
-    [[ -n "$watcher_pid" ]] && wait "$watcher_pid" 2>/dev/null || true
 
     # Recover session log from backup if truncated by subagent stdout redirection
     prizm_recover_session_log "$session_log"
 
-    # Map SIGTERM (143) to timeout code 124
-    if [[ $exit_code -eq 143 ]]; then
-        exit_code=124
-    fi
     _SPAWN_EXIT_CODE="$exit_code"
 
     # Check for stale-kill marker (heartbeat killed the process due to no progress)
     local stale_kill_marker="$session_dir/logs/stale-kill.json"
     local was_stale_killed=false
+    local stale_kill_reason=""
     if [[ -f "$stale_kill_marker" ]]; then
         was_stale_killed=true
-        log_warn "Session was stale-killed by heartbeat monitor (no progress for too long)"
+        stale_kill_reason=$(python3 - "$stale_kill_marker" <<'PY' 2>/dev/null || true
+import json, sys
+try:
+    with open(sys.argv[1], encoding="utf-8") as fh:
+        print(json.load(fh).get("reason") or "")
+except Exception:
+    pass
+PY
+)
+        if [[ "$stale_kill_reason" == "error_loop" ]]; then
+            log_warn "Session was killed by heartbeat monitor due to repeated read-offset/wasted-call errors"
+        else
+            log_warn "Session was stale-killed by heartbeat monitor (no progress for too long)"
+        fi
     fi
 
     local was_infra_error=false
@@ -213,16 +208,17 @@ spawn_and_wait_session() {
         log_warn "Session failed due to structured AI runtime/context error"
         log_warn "AI runtime errors are retried without consuming code retry budget"
         session_status="infra_error"
-    elif [[ $exit_code -eq 124 ]]; then
-        log_warn "Session timed out after ${SESSION_TIMEOUT}s"
-        session_status="timed_out"
     elif [[ "$was_infra_error" == true ]]; then
         log_warn "Session failed due to AI CLI/provider infrastructure error"
         log_warn "Infrastructure errors are retried without consuming code retry budget"
         session_status="infra_error"
     elif [[ "$was_stale_killed" == true ]]; then
-        log_warn "Session stale-killed (no progress for ${STALE_KILL_THRESHOLD}s)"
-        log_warn "Stale-killed sessions are treated as failed; dev branch is preserved for inspection"
+        if [[ "$stale_kill_reason" == "error_loop" ]]; then
+            log_warn "Session killed due to repeated read-offset/wasted-call error loop"
+        else
+            log_warn "Session stale-killed (no progress for ${STALE_KILL_THRESHOLD}s)"
+        fi
+        log_warn "Heartbeat-killed sessions are treated as failed; dev branch is preserved for inspection"
         session_status="crashed"
     elif [[ $exit_code -ne 0 ]]; then
         log_warn "Session exited with code $exit_code"
@@ -240,10 +236,10 @@ spawn_and_wait_session() {
             # No commits found — check if there are uncommitted changes (session
             # did work but didn't commit, e.g. context window exhausted)
             local uncommitted=""
-            uncommitted=$(git -C "$project_root" status --porcelain 2>/dev/null | head -1 || true)
+            uncommitted=$(prizm_git_status_safe "$project_root" | head -1 || true)
             if [[ -n "$uncommitted" ]]; then
                 log_warn "Session exited cleanly but produced no commits (uncommitted changes found) — auto-committing..."
-                git -C "$project_root" add -A 2>/dev/null || true
+                prizm_git_add_all_safe "$project_root" || true
                 if git -C "$project_root" commit --no-verify -m "chore($feature_id): auto-commit session work" 2>/dev/null; then
                     log_info "Auto-commit succeeded"
                     session_status="success"
@@ -262,10 +258,10 @@ spawn_and_wait_session() {
     if [[ "$session_status" == "success" ]]; then
         if git -C "$project_root" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
             local dirty_files=""
-            dirty_files=$(git -C "$project_root" status --porcelain 2>/dev/null || true)
+            dirty_files=$(prizm_git_status_safe "$project_root" || true)
             if [[ -n "$dirty_files" ]]; then
                 log_info "Auto-committing remaining session artifacts..."
-                git -C "$project_root" add -A 2>/dev/null || true
+                prizm_git_add_all_safe "$project_root" || true
                 git -C "$project_root" commit --no-verify --amend --no-edit 2>/dev/null \
                     || git -C "$project_root" commit --no-verify -m "chore($feature_id): include remaining session artifacts" 2>/dev/null \
                     || true
@@ -385,8 +381,9 @@ finalize_feature_status_after_branch_return() {
     local session_id="$3"
     local session_status="$4"
     local max_retries="$5"
-    local session_dir="$6"
-    local base_branch="${7:-main}"
+    local max_infra_retries="$6"
+    local session_dir="$7"
+    local base_branch="${8:-main}"
 
     local feature_slug="${_SPAWN_FEATURE_SLUG:-}"
     local progress_json="$session_dir/logs/progress.json"
@@ -422,6 +419,7 @@ finalize_feature_status_after_branch_return() {
         --session-status "$session_status" \
         --session-id "$session_id" \
         --max-retries "$max_retries" \
+        --max-infra-retries "$max_infra_retries" \
         --action update 2>&1) || {
         log_error "Failed to update feature status: $update_output"
         log_error ".prizmkit/plans/feature-list.json may be out of sync. Manual intervention needed."
@@ -467,6 +465,8 @@ for feat in data.get('features', []):
                 --state-dir "$STATE_DIR" \
                 --feature-id "$_interrupted_id" \
                 --session-status "failed" \
+                --max-retries "$MAX_RETRIES" \
+                --max-infra-retries "$MAX_INFRA_RETRIES" \
                 --action update 2>/dev/null || true
             log_info "Feature $_interrupted_id marked as failed due to interrupt"
         fi
@@ -497,29 +497,6 @@ check_dependencies() {
     prizm_ensure_git_repo "$_project_root"
 }
 
-run_log_cleanup() {
-    if [[ "$LOG_CLEANUP_ENABLED" != "1" ]]; then
-        return 0
-    fi
-
-    local cleanup_result
-    cleanup_result=$(python3 "$SCRIPTS_DIR/cleanup-logs.py" \
-        --state-dir "$STATE_DIR" \
-        --retention-days "$LOG_RETENTION_DAYS" \
-        --max-total-mb "$LOG_MAX_TOTAL_MB" 2>/dev/null) || {
-        log_warn "Log cleanup failed (continuing)"
-        return 0
-    }
-
-    local deleted reclaimed_kb
-    deleted=$(echo "$cleanup_result" | python3 -c "import sys,json; print(json.load(sys.stdin).get('deleted_files', 0))" 2>/dev/null || echo "0")
-    reclaimed_kb=$(echo "$cleanup_result" | python3 -c "import sys,json; print(int(json.load(sys.stdin).get('reclaimed_bytes', 0)/1024))" 2>/dev/null || echo "0")
-
-    if [[ "$deleted" -gt 0 ]]; then
-        log_info "Log cleanup: deleted $deleted files, reclaimed ${reclaimed_kb}KB"
-    fi
-}
-
 # ============================================================
 # run-one: Run a single feature with full control
 # ============================================================
@@ -583,13 +560,14 @@ run_one() {
                 critic_override="false"
                 shift
                 ;;
-            --timeout)
+            --max-infra-retries)
                 shift
                 if [[ $# -eq 0 ]]; then
-                    log_error "--timeout requires a value in seconds"
+                    log_error "--max-infra-retries requires a value"
                     exit 1
                 fi
-                SESSION_TIMEOUT="$1"
+                prizm_require_positive_int "--max-infra-retries" "$1"
+                MAX_INFRA_RETRIES="$1"
                 shift
                 ;;
             F-*|f-*)
@@ -635,7 +613,6 @@ run_one() {
     fi
 
     check_dependencies
-    run_log_cleanup
 
     # Initialize pipeline state if needed (same logic as run_all)
     if [[ ! -f "$STATE_DIR/pipeline.json" ]]; then
@@ -887,11 +864,6 @@ else:
     esac
     log_info "Pipeline mode: ${BOLD}$pipeline_mode${NC} ($_run_one_mode_desc)"
     log_info "Agents: $agent_count (critic: $([ "$critic_enabled" = "true" ] && echo "enabled" || echo "disabled"))"
-    if [[ $SESSION_TIMEOUT -gt 0 ]]; then
-        log_info "Session timeout: ${SESSION_TIMEOUT}s"
-    else
-        log_info "Session timeout: none"
-    fi
     log_info "Prompt: $bootstrap_prompt"
     log_info "Log: $session_dir/logs/session.log"
     echo -e "${BOLD}════════════════════════════════════════════════════${NC}"
@@ -916,6 +888,8 @@ else:
                 --state-dir "$STATE_DIR" \
                 --feature-id "$feature_id" \
                 --session-status "failed" \
+                --max-retries "$MAX_RETRIES" \
+                --max-infra-retries "$MAX_INFRA_RETRIES" \
                 --action update 2>/dev/null || true
             log_info "Feature $feature_id marked as failed due to interrupt"
         fi
@@ -949,12 +923,13 @@ else:
     if branch_create "$_proj_root" "$_branch_name" "$_source_branch"; then
         _DEV_BRANCH_NAME="$_branch_name"
     else
-        log_warn "Failed to create branch; running session on current branch"
+        log_error "Failed to create branch: $_branch_name from $_source_branch — aborting to avoid running on $_source_branch"
+        return 1
     fi
 
     spawn_and_wait_session \
         "$feature_id" "$feature_list" "$session_id" \
-        "$bootstrap_prompt" "$session_dir" 999 "$feature_model" "$_ORIGINAL_BRANCH"
+        "$bootstrap_prompt" "$session_dir" "$MAX_RETRIES" "$feature_model" "$_ORIGINAL_BRANCH"
     local session_status="$_SPAWN_RESULT"
 
     # Merge dev branch back to original on success
@@ -976,12 +951,11 @@ else:
     branch_ensure_return "$_proj_root" "$_ORIGINAL_BRANCH"
 
     finalize_feature_status_after_branch_return \
-        "$feature_id" "$feature_list" "$session_id" "$session_status" 999 "$session_dir" "$_ORIGINAL_BRANCH"
+        "$feature_id" "$feature_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$session_dir" "$_ORIGINAL_BRANCH"
 
     # Commit feature status update on the original branch (after guaranteed return)
     if ! git -C "$_proj_root" diff --quiet "$feature_list" 2>/dev/null; then
-        git -C "$_proj_root" add "$feature_list" 2>/dev/null || true
-        git -C "$_proj_root" commit --no-verify -m "chore($feature_id): update feature status" 2>/dev/null || true
+        prizm_git_commit_paths "$_proj_root" "chore($feature_id): update feature status" "$feature_list"
     fi
 
     echo ""
@@ -1029,7 +1003,6 @@ main() {
     fi
 
     check_dependencies
-    run_log_cleanup
 
     # Initialize pipeline state if needed
     if [[ ! -f "$STATE_DIR/pipeline.json" ]]; then
@@ -1080,12 +1053,8 @@ main() {
     if [[ -n "$features_filter" ]]; then
         log_info "Features filter: $features_filter"
     fi
-    log_info "Max retries per feature: $MAX_RETRIES"
-    if [[ $SESSION_TIMEOUT -gt 0 ]]; then
-        log_info "Session timeout: ${SESSION_TIMEOUT}s"
-    else
-        log_info "Session timeout: none"
-    fi
+    log_info "Max code retries per feature: $MAX_RETRIES"
+    log_info "Max infrastructure retries per feature: $MAX_INFRA_RETRIES"
     log_info "AI CLI: $CLI_CMD (platform: $PLATFORM)"
     if [[ -n "${MODEL:-}" ]]; then
         log_info "Default Model: $MODEL"
@@ -1259,28 +1228,38 @@ DEPLOY_PROMPT_EOF
         fi
 
         # Parse feature info
-        local feature_id feature_title retry_count resume_phase
+        local feature_id feature_title retry_count infra_error_count resume_phase
         feature_id=$(echo "$next_feature" | jq -r '.feature_id')
         feature_title=$(echo "$next_feature" | jq -r '.title')
         retry_count=$(echo "$next_feature" | jq -r '.retry_count // 0')
+        infra_error_count=$(echo "$next_feature" | jq -r '.infra_error_count // 0')
         resume_phase=$(echo "$next_feature" | jq -r '.resume_from_phase // "null"')
 
         echo ""
         echo -e "${BOLD}────────────────────────────────────────────────────${NC}"
         log_info "Feature: ${BOLD}$feature_id${NC} — $feature_title"
-        log_info "Retry: $retry_count / $MAX_RETRIES"
+        log_info "Code retry: $retry_count / $MAX_RETRIES"
+        log_info "Infrastructure retry: $infra_error_count / $MAX_INFRA_RETRIES"
         if [[ "$resume_phase" != "null" ]]; then
             log_info "Resuming from Phase $resume_phase"
         fi
         echo -e "${BOLD}────────────────────────────────────────────────────${NC}"
 
-        # Commit dirty working tree before starting feature
+        # Auto-commit any uncommitted workspace changes before starting the pipeline.
+        # This ensures a clean baseline without blocking execution.
         local _dirty_files=""
-        _dirty_files=$(git -C "$_proj_root" status --porcelain 2>/dev/null || true)
+        _dirty_files=$(prizm_git_status_non_bookkeeping "$_proj_root" "$STATE_DIR" "$feature_list" || true)
         if [[ -n "$_dirty_files" ]]; then
-            log_info "Dirty working tree detected — committing before $feature_id..."
-            git -C "$_proj_root" add -A 2>/dev/null || true
-            git -C "$_proj_root" commit --no-verify -m "ready for run $feature_id" 2>/dev/null || true
+            log_info "Auto-committing workspace changes before $feature_id..."
+            prizm_git_add_all_safe "$_proj_root"
+            git -C "$_proj_root" commit --no-verify -m "chore: workspace snapshot ready for run $feature_id" 2>/dev/null || true
+        fi
+
+        local _bookkeeping_dirty=""
+        _bookkeeping_dirty=$(prizm_git_status_bookkeeping "$_proj_root" "$STATE_DIR" "$feature_list" || true)
+        if [[ -n "$_bookkeeping_dirty" ]]; then
+            log_info "Committing pending pipeline bookkeeping before $feature_id..."
+            prizm_git_commit_paths "$_proj_root" "chore($feature_id): include pipeline bookkeeping" "$feature_list" "$STATE_DIR"
         fi
 
         # Mark feature as in-progress BEFORE creating dev branch
@@ -1293,8 +1272,7 @@ DEPLOY_PROMPT_EOF
             --action start >/dev/null 2>&1 || true
         # Commit the in_progress status on the original branch
         if ! git -C "$_proj_root" diff --quiet "$feature_list" 2>/dev/null; then
-            git -C "$_proj_root" add "$feature_list" 2>/dev/null || true
-            git -C "$_proj_root" commit --no-verify -m "chore($feature_id): mark in_progress" 2>/dev/null || true
+            prizm_git_commit_paths "$_proj_root" "chore($feature_id): mark in_progress" "$feature_list"
         fi
 
         # Create per-feature dev branch (from the now-updated original branch)
@@ -1303,8 +1281,9 @@ DEPLOY_PROMPT_EOF
             _DEV_BRANCH_NAME="$_feature_branch"
             log_info "Dev branch: $_feature_branch"
         else
-            log_warn "Failed to create dev branch; running on current branch: $_ORIGINAL_BRANCH"
+            log_error "Failed to create dev branch: $_feature_branch from $_ORIGINAL_BRANCH — aborting to avoid running on $_ORIGINAL_BRANCH"
             _DEV_BRANCH_NAME=""
+            break
         fi
 
         # Generate session ID and bootstrap prompt
@@ -1397,20 +1376,20 @@ DEPLOY_PROMPT_EOF
         branch_ensure_return "$_proj_root" "$_ORIGINAL_BRANCH"
 
         finalize_feature_status_after_branch_return \
-            "$feature_id" "$feature_list" "$session_id" "$session_status" "$MAX_RETRIES" "$session_dir" "$_ORIGINAL_BRANCH"
+            "$feature_id" "$feature_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$session_dir" "$_ORIGINAL_BRANCH"
         local item_status_after_session="${_SPAWN_ITEM_STATUS:-}"
 
         # Commit feature status update on the original branch (after guaranteed return)
         if ! git -C "$_proj_root" diff --quiet "$feature_list" 2>/dev/null; then
-            git -C "$_proj_root" add "$feature_list" 2>/dev/null || true
-            git -C "$_proj_root" commit --no-verify -m "chore($feature_id): update feature status" 2>/dev/null || true
+            prizm_git_commit_paths "$_proj_root" "chore($feature_id): update feature status" "$feature_list"
         fi
 
         session_count=$((session_count + 1))
         total_subagent_calls=$((total_subagent_calls + _SUBAGENT_COUNT))
 
         # Stop-on-failure: abort only after the task is actually marked failed.
-        # Pending retry outcomes, including infrastructure errors, keep running.
+        # Pending retry outcomes keep running; both code and infra failures have
+        # bounded budgets in update-feature-status.py.
         if [[ "$session_status" != "success" && "$STOP_ON_FAILURE" == "1" && "$item_status_after_session" == "failed" ]]; then
             echo ""
             log_error "════════════════════════════════════════════════════"
@@ -1420,7 +1399,7 @@ DEPLOY_PROMPT_EOF
             log_error "════════════════════════════════════════════════════"
             break
         elif [[ "$session_status" != "success" && "$STOP_ON_FAILURE" == "1" ]]; then
-            log_info "STOP_ON_FAILURE: $feature_id is ${item_status_after_session:-unknown}; retry budget not exhausted, continuing."
+            log_info "STOP_ON_FAILURE: $feature_id is ${item_status_after_session:-unknown}; code/infra retry budget not exhausted, continuing."
         fi
 
         # Brief pause before next iteration
@@ -1458,20 +1437,20 @@ show_help() {
     echo "  --no-critic                 Disable critic review (overrides feature-list setting)"
     echo "  --clean                     Delete artifacts and reset before running"
     echo "  --no-reset                  Skip feature status reset step"
-    echo "  --timeout N                 Session timeout in seconds (default: 0 = no limit)"
+    echo "  --max-infra-retries N       Max infrastructure/provider retries (default: 3)"
+    echo ""
+    echo "Pipeline Options (run all):"
+    echo "  --max-infra-retries N       Max infrastructure/provider retries (default: 3)"
     echo ""
     echo "Environment Variables:"
-    echo "  MAX_RETRIES           Max retries per feature (default: 3)"
-    echo "  SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)"
+    echo "  MAX_RETRIES           Max code retries per feature (default: 3)"
+    echo "  MAX_INFRA_RETRIES     Max infrastructure/provider retries per feature (default: 3)"
     echo "  AI_CLI                AI CLI command name (auto-detected: cbc, claude, or codex)"
     echo "  MODEL                 AI model ID (e.g. claude-opus-4.6, claude-sonnet-4.6, claude-haiku-4.5)"
     echo "  PRIZMKIT_EFFORT       AI reasoning effort (low|medium|high|xhigh|max; max is Claude Code only)"
     echo "  HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)"
     echo "  STALE_KILL_THRESHOLD  Auto-kill session after N seconds of no progress (default: 900)"
     echo "  HEARTBEAT_STALE_THRESHOLD  Heartbeat stale threshold in seconds (default: 600)"
-    echo "  LOG_CLEANUP_ENABLED   Run log cleanup before execution (default: 1)"
-    echo "  LOG_RETENTION_DAYS    Delete logs older than N days (default: 14)"
-    echo "  LOG_MAX_TOTAL_MB      Keep total logs under N MB (default: 1024)"
     echo "  PIPELINE_MODE         Override mode for all features: lite|standard|full"
     echo "  ENABLE_CRITIC         Enable critic review for all features: true|false"
     echo "  STOP_ON_FAILURE       Stop pipeline when a task exhausts retries: 0|1 (default: 0)"
@@ -1483,10 +1462,8 @@ show_help() {
     echo "  ./run-feature.sh run F-007 --dry-run                          # Inspect generated prompt"
     echo "  ./run-feature.sh run F-007 --dry-run --mode lite               # Test lite mode"
     echo "  ./run-feature.sh run F-007 --resume-phase 6                    # Skip to implementation"
-    echo "  ./run-feature.sh run F-007 --mode full --timeout 3600          # Full mode, 1h timeout"
     echo "  ./run-feature.sh run F-007 --clean --mode standard             # Clean + run standard"
     echo "  ./run-feature.sh status                                        # Show pipeline status"
-    echo "  MAX_RETRIES=5 SESSION_TIMEOUT=7200 ./run-feature.sh run        # Custom config"
     echo "  MODEL=claude-sonnet-4.6 ./run-feature.sh run                    # Use Sonnet model"
     echo "  MODEL=claude-haiku-4.5 ./run-feature.sh test-cli                # Test with Haiku"
 }
@@ -1512,6 +1489,16 @@ case "${1:-run}" in
                         _run_features_filter="$1"
                         shift
                         ;;
+                    --max-infra-retries)
+                        shift
+                        if [[ $# -eq 0 ]]; then
+                            log_error "--max-infra-retries requires a value"
+                            exit 1
+                        fi
+                        prizm_require_positive_int "--max-infra-retries" "$1"
+                        MAX_INFRA_RETRIES="$1"
+                        shift
+                        ;;
                     *)
                         _run_feature_list="$1"
                         shift

package/bundled/dev-pipeline/run-recovery.sh

@@ -23,7 +23,6 @@ set -euo pipefail
 #   ./run-recovery.sh help                  Show help
 #
 # Environment Variables:
-#   SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)
 #   AI_CLI                AI CLI command name (auto-detected: cbc, claude, or codex)
 #   PRIZMKIT_PLATFORM     Platform hint for custom AI_CLI wrappers: codebuddy, claude, or codex
 #   MODEL                 AI model to use (e.g. claude-opus-4.6)
@@ -62,7 +61,6 @@ if [[ -z "$RECOVERY_DETECT_SCRIPT" ]]; then
 fi
 
 # Configuration
-SESSION_TIMEOUT=${SESSION_TIMEOUT:-0}
 HEARTBEAT_INTERVAL=${HEARTBEAT_INTERVAL:-30}
 STALE_KILL_THRESHOLD=${STALE_KILL_THRESHOLD:-900}
 VERBOSE=${VERBOSE:-0}
@@ -108,7 +106,6 @@ Options (for 'run' command):
   --model M   Override AI model (e.g. claude-opus-4.6)
 
 Environment Variables:
-  SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)
   MODEL                 AI model to use
   PRIZMKIT_EFFORT       AI reasoning effort (low|medium|high|xhigh|max; max is Claude Code only)
   VERBOSE               Set to 1 for verbose AI CLI output
@@ -283,13 +280,6 @@ spawn_recovery_session() {
     start_progress_parser "$session_log" "$progress_json" "$SCRIPTS_DIR"
     local parser_pid="${_PARSER_PID:-}"
 
-    # Timeout watchdog
-    local watcher_pid=""
-    if [[ $SESSION_TIMEOUT -gt 0 ]]; then
-        ( sleep "$SESSION_TIMEOUT" && kill -TERM "$_SESSION_PID" 2>/dev/null ) &
-        watcher_pid=$!
-    fi
-
     # Heartbeat monitor
     start_heartbeat "$_SESSION_PID" "$session_log" "$progress_json" "$HEARTBEAT_INTERVAL" "$STALE_KILL_THRESHOLD"
     _HEARTBEAT_PID="${_HEARTBEAT_PID:-}"
@@ -303,17 +293,10 @@ spawn_recovery_session() {
     fi
     _SESSION_PID=""
 
-    # Cleanup background processes
-    [[ -n "$watcher_pid" ]] && kill "$watcher_pid" 2>/dev/null || true
+    # Clean up heartbeat and parser
     stop_heartbeat "${_HEARTBEAT_PID:-}"
     _HEARTBEAT_PID=""
     stop_progress_parser "$parser_pid"
-    [[ -n "$watcher_pid" ]] && wait "$watcher_pid" 2>/dev/null || true
-
-    # Map SIGTERM to timeout
-    if [[ $exit_code -eq 143 ]]; then
-        exit_code=124
-    fi
 
     # Check for stale-kill
     local stale_kill_marker="$session_dir/logs/stale-kill.json"
@@ -347,12 +330,6 @@ detect_session_outcome() {
     local was_stale_killed=false
     [[ -f "$stale_kill_marker" ]] && was_stale_killed=true
 
-    if [[ $exit_code -eq 124 ]]; then
-        log_warn "Session timed out"
-        echo "timed_out"
-        return
-    fi
-
     if [[ "$was_stale_killed" == true ]]; then
         log_warn "Session stale-killed"
     fi
@@ -611,14 +588,6 @@ cmd_run() {
             echo "    • Commit manually: git add <files> && git commit"
             echo ""
             ;;
-        timed_out)
-            log_warn "Session timed out before completing."
-            echo ""
-            echo "  Options:"
-            echo "    • Run recovery again: ./run-recovery.sh"
-            echo "    • Increase timeout: SESSION_TIMEOUT=3600 ./run-recovery.sh"
-            echo ""
-            ;;
         no_changes)
             log_warn "Session completed but produced no changes."
             echo ""