prizmkit 1.1.98 → 1.1.100

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

@@ -23,8 +23,7 @@ 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)
-#   MAX_LOG_SIZE         Session log size threshold for compact warning (default: 2097152 = 2MB, set 0 to disable)
-#   LOG_MONITOR_POLL_INTERVAL  Log monitor poll interval in seconds (default: 30)
+#   MAX_LOG_SIZE         Include passive log-size checkpoint guidance above this threshold (default: 2097152 = 2MB, set 0 to disable)
 #   DEV_BRANCH            Custom dev branch name (default: auto-generated refactor/pipeline-{run_id})
 #   AUTO_PUSH             Auto-push to remote after successful refactor (default: 0). Set to 1 to enable.
 #   STOP_ON_FAILURE       Stop pipeline after a task exhausts all retries (default: 0). Set to 1 to stop.
@@ -140,6 +139,16 @@ spawn_and_wait_session() {
         log_warn "Session was stale-killed by heartbeat monitor (no progress for too long)"
     fi
 
+    local was_context_overflow=false
+    if prizm_detect_context_overflow_error "$session_log" "$progress_json"; then
+        was_context_overflow=true
+    fi
+
+    local was_ai_runtime_error=false
+    if prizm_detect_ai_runtime_error "$session_log" "$progress_json"; then
+        was_ai_runtime_error=true
+    fi
+
     local was_infra_error=false
     if [[ $exit_code -ne 0 ]] && prizm_detect_infra_error "$session_log" "$progress_json"; then
         was_infra_error=true
@@ -163,7 +172,22 @@ spawn_and_wait_session() {
     project_root="$PROJECT_ROOT"
     local default_branch="$base_branch"
 
-    if [[ "$was_infra_error" == true ]]; then
+    local has_commits=""
+    if git -C "$project_root" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
+        has_commits=$(git -C "$project_root" log "${default_branch}..HEAD" --oneline 2>/dev/null | head -1)
+    fi
+
+    if [[ -n "$has_commits" ]]; then
+        session_status="success"
+    elif [[ "$was_context_overflow" == true ]]; then
+        log_warn "Session failed because the AI CLI exceeded the model context window"
+        log_warn "Context overflow is retried without consuming code retry budget"
+        session_status="context_overflow"
+    elif [[ "$was_ai_runtime_error" == true ]]; then
+        log_warn "Session failed due to structured AI runtime error"
+        log_warn "AI runtime errors are retried without consuming code retry budget"
+        session_status="infra_error"
+    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"
@@ -299,6 +323,12 @@ finalize_refactor_status_after_branch_return() {
     local session_status="$4"
     local max_retries="$5"
     local max_infra_retries="$6"
+    local base_branch="${7:-main}"
+    local active_dev_branch="${8:-}"
+    local fatal_error_code="${9:-}"
+    local continuation_summary_path="${10:-}"
+    local no_progress_count="${11:-0}"
+    local progress_fingerprint="${12:-}"
 
     # Update refactor status on the original branch. The caller commits the
     # resulting refactor-list diff immediately after this helper returns.
@@ -311,6 +341,12 @@ finalize_refactor_status_after_branch_return() {
         --session-id "$session_id" \
         --max-retries "$max_retries" \
         --max-infra-retries "$max_infra_retries" \
+        --active-dev-branch "$active_dev_branch" \
+        --base-branch "$base_branch" \
+        --last-fatal-error-code "$fatal_error_code" \
+        --continuation-summary-path "$continuation_summary_path" \
+        --no-progress-count "$no_progress_count" \
+        --progress-fingerprint "$progress_fingerprint" \
         --action update) || {
         log_error "Failed to update refactor status: $update_output"
         update_output=""
@@ -544,7 +580,7 @@ sys.exit(1)
                 log_info "Removed $dev_team_dir"
             fi
 
-            local refactor_state_dir="$STATE_DIR/refactors/$refactor_id"
+            local refactor_state_dir="$STATE_DIR/$refactor_id"
             if [[ -d "$refactor_state_dir" ]]; then
                 rm -rf "$refactor_state_dir"
                 log_info "Removed $refactor_state_dir"
@@ -569,7 +605,7 @@ sys.exit(1)
     local run_id session_id session_dir bootstrap_prompt
     run_id=$(jq -r '.run_id' "$STATE_DIR/pipeline.json")
     session_id="${refactor_id}-$(date +%Y%m%d%H%M%S)"
-    session_dir="$STATE_DIR/refactors/$refactor_id/sessions/$session_id"
+    session_dir="$STATE_DIR/$refactor_id/sessions/$session_id"
     mkdir -p "$session_dir/logs"
 
     bootstrap_prompt="$session_dir/bootstrap-prompt.md"
@@ -578,7 +614,7 @@ sys.exit(1)
     local retry_count
     retry_count=$(python3 -c "
 import json, os
-status_path = os.path.join('$STATE_DIR', 'refactors', '$refactor_id', 'status.json')
+status_path = os.path.join('$STATE_DIR', '$refactor_id', 'status.json')
 if os.path.isfile(status_path):
     with open(status_path) as f:
         d = json.load(f)
@@ -589,7 +625,7 @@ else:
     local resume_phase
     resume_phase=$(python3 -c "
 import json, os
-status_path = os.path.join('$STATE_DIR', 'refactors', '$refactor_id', 'status.json')
+status_path = os.path.join('$STATE_DIR', '$refactor_id', 'status.json')
 if os.path.isfile(status_path):
     with open(status_path) as f:
         d = json.load(f)
@@ -751,6 +787,11 @@ else:
         "$bootstrap_prompt" "$session_dir" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$refactor_model" "$_ORIGINAL_BRANCH"
     local session_status="$_SPAWN_RESULT"
 
+    local active_dev_branch_for_status="$_DEV_BRANCH_NAME"
+    local continuation_summary_path=""
+    local no_progress_count=0
+    local progress_after=""
+
     # Merge dev branch back to original on success
     if [[ "$session_status" == "success" && -n "$_DEV_BRANCH_NAME" ]]; then
         if branch_merge "$_proj_root" "$_DEV_BRANCH_NAME" "$_ORIGINAL_BRANCH" "$AUTO_PUSH"; then
@@ -770,7 +811,7 @@ else:
     branch_ensure_return "$_proj_root" "$_ORIGINAL_BRANCH"
 
     finalize_refactor_status_after_branch_return \
-        "$refactor_id" "$refactor_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES"
+        "$refactor_id" "$refactor_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$_ORIGINAL_BRANCH" "$active_dev_branch_for_status" "context_overflow" "$continuation_summary_path" "$no_progress_count" "${progress_after:-}"
 
     # Commit refactor status update on the original branch (after guaranteed return)
     if ! git -C "$_proj_root" diff --quiet "$refactor_list" 2>/dev/null; then
@@ -1024,13 +1065,21 @@ DEPLOY_PROMPT_EOF
         fi
 
         # Parse refactor info
-        local refactor_id refactor_title refactor_complexity retry_count infra_error_count resume_phase
+        local refactor_id refactor_title refactor_complexity retry_count infra_error_count resume_phase active_dev_branch base_branch continuation_pending continuation_reason context_overflow_count continuation_count previous_session_id no_progress_count
         refactor_id=$(echo "$next_refactor" | jq -r '.refactor_id')
         refactor_title=$(echo "$next_refactor" | jq -r '.title')
         refactor_complexity=$(echo "$next_refactor" | jq -r '.complexity')
         retry_count=$(echo "$next_refactor" | jq -r '.retry_count // 0')
         infra_error_count=$(echo "$next_refactor" | jq -r '.infra_error_count // 0')
         resume_phase=$(echo "$next_refactor" | jq -r '.resume_from_phase // "null"')
+        active_dev_branch=$(echo "$next_refactor" | jq -r '.active_dev_branch // ""')
+        base_branch=$(echo "$next_refactor" | jq -r '.base_branch // ""')
+        continuation_pending=$(echo "$next_refactor" | jq -r '.continuation_pending // false')
+        continuation_reason=$(echo "$next_refactor" | jq -r '.continuation_reason // ""')
+        context_overflow_count=$(echo "$next_refactor" | jq -r '.context_overflow_count // 0')
+        continuation_count=$(echo "$next_refactor" | jq -r '.continuation_count // 0')
+        previous_session_id=$(echo "$next_refactor" | jq -r '.last_context_overflow_session_id // ""')
+        no_progress_count=$(echo "$next_refactor" | jq -r '.no_progress_count // 0')
 
         echo ""
         echo -e "${BOLD}────────────────────────────────────────────────────${NC}"
@@ -1041,6 +1090,10 @@ DEPLOY_PROMPT_EOF
         if [[ "$resume_phase" != "null" ]]; then
             log_info "Resuming from Phase $resume_phase"
         fi
+        if [[ "$continuation_pending" == "true" && -n "$active_dev_branch" ]]; then
+            log_info "Continuation pending (${continuation_reason:-context_overflow}); reusing dev branch: $active_dev_branch"
+            log_info "Continuation count: $continuation_count (context overflows: $context_overflow_count)"
+        fi
         echo -e "${BOLD}────────────────────────────────────────────────────${NC}"
 
         # Auto-commit any uncommitted workspace changes before starting the pipeline.
@@ -1073,8 +1126,20 @@ DEPLOY_PROMPT_EOF
             prizm_git_commit_paths "$_proj_root" "chore($refactor_id): mark in_progress" "$refactor_list"
         fi
 
-        # Create per-refactor dev branch (from the now-updated original branch)
-        local _refactor_branch="${DEV_BRANCH:-refactor/${refactor_id}-$(date +%Y%m%d%H%M)}"
+        # Continuation base branch from runtime state is authoritative for branch reuse.
+        if [[ "$continuation_pending" == "true" && -n "$base_branch" ]]; then
+            _ORIGINAL_BRANCH="$base_branch"
+        fi
+
+        # Create or reuse per-refactor dev branch (from the now-updated original branch)
+        local _refactor_branch
+        if [[ -n "$DEV_BRANCH" ]]; then
+            _refactor_branch="$DEV_BRANCH"
+        elif [[ "$continuation_pending" == "true" && -n "$active_dev_branch" ]]; then
+            _refactor_branch="$active_dev_branch"
+        else
+            _refactor_branch="refactor/${refactor_id}-$(date +%Y%m%d%H%M)"
+        fi
         if branch_create "$_proj_root" "$_refactor_branch" "$_ORIGINAL_BRANCH"; then
             _DEV_BRANCH_NAME="$_refactor_branch"
             log_info "Dev branch: $_refactor_branch"
@@ -1084,15 +1149,17 @@ DEPLOY_PROMPT_EOF
             break
         fi
 
+        while true; do
         # Generate session
         local session_id run_id
         run_id=$(jq -r '.run_id' "$STATE_DIR/pipeline.json")
         session_id="${refactor_id}-$(date +%Y%m%d%H%M%S)"
 
-        local session_dir="$STATE_DIR/refactors/$refactor_id/sessions/$session_id"
+        local session_dir="$STATE_DIR/$refactor_id/sessions/$session_id"
         mkdir -p "$session_dir/logs"
 
         local bootstrap_prompt="$session_dir/bootstrap-prompt.md"
+        local continuation_summary_path=".prizmkit/refactor/${refactor_id}/continuation-summary.md"
 
         local main_prompt_args=(
             --refactor-list "$refactor_list"
@@ -1105,6 +1172,19 @@ DEPLOY_PROMPT_EOF
             --output "$bootstrap_prompt"
         )
 
+        if [[ "$continuation_pending" == "true" ]]; then
+            main_prompt_args+=(
+                --continuation-mode "${continuation_reason:-context_overflow}"
+                --previous-session-id "$previous_session_id"
+                --continuation-count "$continuation_count"
+                --context-overflow-count "$context_overflow_count"
+                --task-type "refactor"
+                --active-dev-branch "$active_dev_branch"
+                --base-branch "$base_branch"
+                --continuation-summary-path "$continuation_summary_path"
+            )
+        fi
+
         # Support PIPELINE_MODE env var (set by launch-refactor-daemon.sh --mode)
         if [[ -n "${PIPELINE_MODE:-}" ]]; then
             main_prompt_args+=(--mode "$PIPELINE_MODE")
@@ -1122,7 +1202,7 @@ DEPLOY_PROMPT_EOF
             log_error "Failed to generate bootstrap prompt for $refactor_id"
             branch_ensure_return "$_proj_root" "$_ORIGINAL_BRANCH"
             _DEV_BRANCH_NAME=""
-            continue
+            continue 2
         }
         local refactor_model pipeline_mode agent_count critic_enabled
         refactor_model=$(echo "$gen_output" | python3 -c "import json,sys; print(json.load(sys.stdin).get('model',''))" 2>/dev/null || echo "")
@@ -1145,6 +1225,11 @@ DEPLOY_PROMPT_EOF
             log_info "Refactor model: $refactor_model"
         fi
 
+        local continuation_artifact_dir=".prizmkit/refactor/${refactor_id}"
+        local continuation_checkpoint_path="$continuation_artifact_dir/workflow-checkpoint.json"
+        local progress_before progress_after
+        progress_before=$(prizm_context_progress_fingerprint "$_proj_root" "refactor" "$refactor_id" "$continuation_artifact_dir" "$continuation_checkpoint_path")
+
         # Spawn session
         log_info "Spawning AI CLI session: $session_id"
         _SPAWN_RESULT=""
@@ -1167,7 +1252,37 @@ DEPLOY_PROMPT_EOF
         local session_status="$_SPAWN_RESULT"
 
         # Merge per-refactor dev branch back to original on success
-        if [[ "$session_status" == "success" && -n "$_DEV_BRANCH_NAME" ]]; then
+        local active_dev_branch_for_status="$_DEV_BRANCH_NAME"
+        if [[ "$session_status" == "context_overflow" ]]; then
+            progress_after=$(prizm_context_progress_fingerprint "$_proj_root" "refactor" "$refactor_id" "$continuation_artifact_dir" "$continuation_checkpoint_path")
+            if prizm_context_progress_changed "$progress_before" "$progress_after"; then
+                no_progress_count=0
+                log_info "Context-overflow session made observable progress; continuing on $_DEV_BRANCH_NAME"
+            else
+                no_progress_count=$((no_progress_count + 1))
+                log_warn "Context-overflow session made no observable progress ($no_progress_count consecutive)"
+            fi
+            if [[ "$no_progress_count" -ge 2 ]]; then
+                log_warn "Pausing automatic continuation: context_overflow_without_git_checkpoint_or_artifact_progress"
+                session_status="stalled_context_continuation"
+            else
+                finalize_refactor_status_after_branch_return \
+                    "$refactor_id" "$refactor_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$_ORIGINAL_BRANCH" "$active_dev_branch_for_status" "context_overflow" "$continuation_summary_path" "$no_progress_count" "$progress_after"
+                session_count=$((session_count + 1))
+                total_subagent_calls=$((total_subagent_calls + _SUBAGENT_COUNT))
+                continuation_pending="true"
+                continuation_reason="context_overflow"
+                previous_session_id="$session_id"
+                active_dev_branch="$_DEV_BRANCH_NAME"
+                base_branch="$_ORIGINAL_BRANCH"
+                context_overflow_count=$((context_overflow_count + 1))
+                continuation_count=$((continuation_count + 1))
+                log_info "Launching automatic continuation for $refactor_id on same branch/worktree"
+                continue
+            fi
+        fi
+
+    if [[ "$session_status" == "success" && -n "$_DEV_BRANCH_NAME" ]]; then
             if branch_merge "$_proj_root" "$_DEV_BRANCH_NAME" "$_ORIGINAL_BRANCH" "$AUTO_PUSH"; then
                 _DEV_BRANCH_NAME=""
             else
@@ -1178,14 +1293,13 @@ DEPLOY_PROMPT_EOF
         elif [[ -n "$_DEV_BRANCH_NAME" ]]; then
             # Session failed — preserve dev branch for inspection
             log_warn "Session failed — dev branch preserved for inspection: $_DEV_BRANCH_NAME"
-            _DEV_BRANCH_NAME=""
         fi
 
         # GUARANTEED: always return to original branch regardless of success/failure/merge outcome
         branch_ensure_return "$_proj_root" "$_ORIGINAL_BRANCH"
 
         finalize_refactor_status_after_branch_return \
-            "$refactor_id" "$refactor_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES"
+            "$refactor_id" "$refactor_list" "$session_id" "$session_status" "$MAX_RETRIES" "$MAX_INFRA_RETRIES" "$_ORIGINAL_BRANCH" "$active_dev_branch_for_status" "context_overflow" "$continuation_summary_path" "$no_progress_count" "${progress_after:-}"
         local item_status_after_session="${_SPAWN_ITEM_STATUS:-}"
 
         # Commit refactor status update on the original branch (after guaranteed return)
@@ -1205,6 +1319,8 @@ DEPLOY_PROMPT_EOF
 
         session_count=$((session_count + 1))
         total_subagent_calls=$((total_subagent_calls + _SUBAGENT_COUNT))
+        break
+        done
 
         # Stop-on-failure: abort only after the task is actually marked failed.
         # Pending retry outcomes, including infrastructure errors, keep running.
@@ -1263,8 +1379,7 @@ show_help() {
     echo "  STRICT_BEHAVIOR_CHECK    Force full test suite after each refactor (default: 1)"
     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 "  MAX_LOG_SIZE             Session log size threshold for compact warning (default: 2097152 = 2MB, set 0 to disable)"
-    echo "  LOG_MONITOR_POLL_INTERVAL  Log monitor poll interval in seconds (default: 30)"
+    echo "  MAX_LOG_SIZE             Include passive log-size checkpoint guidance above this threshold (default: 2097152 = 2MB, set 0 to disable)"
     echo "  STOP_ON_FAILURE          Stop pipeline when a task exhausts retries: 0|1 (default: 0)"
     echo ""
     echo "Examples:"

package/bundled/dev-pipeline/scripts/check-session-status.py

@@ -22,6 +22,77 @@ ID_FIELDS = ["feature_id", "bug_id", "refactor_id"]
 
 LOGGER = setup_logging("check-session-status")
 
+CONTEXT_OVERFLOW_CODE = "context_overflow"
+CONTEXT_CODE_VALUES = {
+    "context_overflow",
+    "context_too_large",
+    "context_length_exceeded",
+    "model_context_window_exceeded",
+}
+CONTEXT_TEXT_PATTERNS = (
+    "input exceeds the context window",
+    "context window of this model",
+    "context window exceeded",
+    "maximum context length exceeded",
+    "prompt is too long",
+    "too many tokens",
+    "input token count exceeds",
+)
+REQUEST_TOO_LARGE_TEXT = "request too large"
+REQUEST_TOO_LARGE_SEMANTICS = ("context", "token", "tokens", "prompt", "input", "model")
+ERROR_LIKE_STATUSES = {"failed", "crashed", "timed_out", "context_overflow"}
+
+
+def _flatten_values(value):
+    if isinstance(value, dict):
+        values = []
+        for child in value.values():
+            values.extend(_flatten_values(child))
+        return values
+    if isinstance(value, list):
+        values = []
+        for child in value:
+            values.extend(_flatten_values(child))
+        return values
+    if value is None:
+        return []
+    return [str(value)]
+
+
+def _is_error_like_status(data):
+    status = str(data.get("status", "")).lower()
+    if status in ERROR_LIKE_STATUSES:
+        return True
+    if data.get("last_result_is_error") is True or data.get("is_error") is True:
+        return True
+    if data.get("api_error_status") not in (None, ""):
+        return True
+    errors = data.get("errors")
+    return isinstance(errors, list) and len(errors) > 0
+
+
+def has_context_overflow_signal(data):
+    """Return true when session-status data carries a context overflow signal.
+
+    Success remains success: direct status handling happens before this helper.
+    Exact machine codes are accepted only from structured code fields. Natural
+    language context/token phrases require a failed/error-like status or an
+    explicit error field to avoid prompt/documentation false positives.
+    """
+    if not isinstance(data, dict):
+        return False
+    for field in ("fatal_error_code", "api_error_code", "error_code", "stop_reason"):
+        value = data.get(field)
+        if isinstance(value, str) and value.lower() in CONTEXT_CODE_VALUES:
+            return True
+    if not _is_error_like_status(data):
+        return False
+    text = " ".join(_flatten_values(data)).lower()
+    if any(pattern in text for pattern in CONTEXT_TEXT_PATTERNS):
+        return True
+    return REQUEST_TOO_LARGE_TEXT in text and any(
+        semantic in text for semantic in REQUEST_TOO_LARGE_SEMANTICS
+    )
 
 def parse_args():
     parser = argparse.ArgumentParser(
@@ -79,12 +150,14 @@ def determine_status(data):
     """Determine the single-line status string from the parsed data.
 
     Returns one of: success, partial_resumable, partial_not_resumable,
-    failed, commit_missing, docs_missing, merge_conflict.
+    failed, commit_missing, docs_missing, merge_conflict, context_overflow.
     """
     status = data.get("status", "")
 
     if status == "success":
         return "success"
+    if status == "context_overflow" or has_context_overflow_signal(data):
+        return "context_overflow"
     elif status == "partial":
         can_resume = data.get("can_resume", False)
         if can_resume: