prizmkit 1.1.91 → 1.1.92

package/bundled/dev-pipeline/scripts/update-feature-status.py

@@ -17,7 +17,7 @@ Usage:
         --feature-list <path> --state-dir <path> \
         --action <get_next|start|update|status|pause|reset|clean|complete|unskip> \
         [--feature-id <id>] [--session-status <status>] \
-        [--session-id <id>] [--max-retries <n>] \
+        [--session-id <id>] [--max-retries <n>] [--max-infra-retries <n>] \
         [--features <filter>]
 """
 
@@ -94,7 +94,13 @@ def parse_args():
         "--max-retries",
         type=int,
         default=3,
-        help="Maximum retry count before marking as failed (default: 3)",
+        help="Maximum code retry count before marking as failed (default: 3)",
+    )
+    parser.add_argument(
+        "--max-infra-retries",
+        type=int,
+        default=3,
+        help="Maximum infrastructure retry count before marking as failed (default: 3)",
     )
     parser.add_argument(
         "--feature-slug",
@@ -170,6 +176,9 @@ def load_feature_status(state_dir, feature_id):
             "feature_id": feature_id,
             "retry_count": 0,
             "max_retries": 3,
+            "max_infra_retries": 3,
+            "infra_error_count": 0,
+            "last_infra_error_session_id": None,
             "sessions": [],
             "last_session_id": None,
             "resume_from_phase": None,
@@ -183,6 +192,9 @@ def load_feature_status(state_dir, feature_id):
             "feature_id": feature_id,
             "retry_count": 0,
             "max_retries": 3,
+            "max_infra_retries": 3,
+            "infra_error_count": 0,
+            "last_infra_error_session_id": None,
             "sessions": [],
             "last_session_id": None,
             "resume_from_phase": None,
@@ -565,6 +577,7 @@ def action_get_next(feature_list_data, state_dir, feature_filter=None):
         "feature_id": chosen_id,
         "title": chosen.get("title", ""),
         "retry_count": chosen_status_data.get("retry_count", 0),
+        "infra_error_count": chosen_status_data.get("infra_error_count", 0),
         "resume_from_phase": chosen_status_data.get("resume_from_phase", None),
     }
     print(json.dumps(result, indent=2, ensure_ascii=False))
@@ -585,6 +598,7 @@ def action_update(args, feature_list_path, state_dir):
     session_status = args.session_status
     session_id = args.session_id
     max_retries = args.max_retries
+    max_infra_retries = args.max_infra_retries
 
     if not feature_id:
         error_out("--feature-id is required for 'update' action")
@@ -600,6 +614,8 @@ def action_update(args, feature_list_path, state_dir):
     new_status = current_list_status
 
     if session_status == "success":
+        fs["infra_error_count"] = 0
+        fs["last_infra_error_session_id"] = None
         # No-op guard: if this exact successful session was already recorded,
         # avoid rewriting state files again (prevents post-commit dirty changes).
         existing_sessions = fs.get("sessions", [])
@@ -616,6 +632,8 @@ def action_update(args, feature_list_path, state_dir):
                 "session_status": session_status,
                 "new_status": "completed",
                 "retry_count": fs.get("retry_count", 0),
+                "infra_error_count": fs.get("infra_error_count", 0),
+                "max_infra_retries": max_infra_retries,
                 "resume_from_phase": fs.get("resume_from_phase"),
                 "updated_at": fs.get("updated_at"),
                 "no_op": True,
@@ -654,15 +672,24 @@ def action_update(args, feature_list_path, state_dir):
             return
     elif session_status == "infra_error":
         # AI CLI/provider outage, auth failure, gateway error, etc.
-        # This is outside the code's control, so keep the item pending without
-        # consuming the task's retry budget.
-        new_status = "pending"
-        fs["infra_error_count"] = fs.get("infra_error_count", 0) + 1
+        # Infra failures do not consume the code retry budget, but they still
+        # need their own bounded budget so a flaky provider cannot loop forever.
+        infra_error_count = fs.get("infra_error_count", 0) + 1
+        fs["infra_error_count"] = infra_error_count
         fs["last_infra_error_session_id"] = session_id
+        fs["max_infra_retries"] = max_infra_retries
+        fs["degraded_reason"] = "infra_error"
         if session_id:
             fs["last_session_id"] = session_id
         fs["resume_from_phase"] = None
 
+        if infra_error_count >= max_infra_retries:
+            new_status = "failed"
+            if session_id:
+                fs["last_failed_session_id"] = session_id
+        else:
+            new_status = "pending"
+
         err = update_feature_in_list(feature_list_path, feature_id, new_status)
         if err:
             error_out("Failed to update .prizmkit/plans/feature-list.json: {}".format(err))
@@ -714,6 +741,8 @@ def action_update(args, feature_list_path, state_dir):
         "session_status": session_status,
         "new_status": new_status,
         "retry_count": fs["retry_count"],
+        "infra_error_count": fs.get("infra_error_count", 0),
+        "max_infra_retries": max_infra_retries,
         "resume_from_phase": fs.get("resume_from_phase"),
         "updated_at": fs["updated_at"],
     }
@@ -1143,6 +1172,8 @@ def action_start(args, feature_list_path, state_dir):
         "feature_id": feature_id,
         "old_status": old_status,
         "new_status": "in_progress",
+        "retry_count": fs.get("retry_count", 0),
+        "infra_error_count": fs.get("infra_error_count", 0),
         "updated_at": fs["updated_at"],
     }
     print(json.dumps(result, indent=2, ensure_ascii=False))
@@ -1171,6 +1202,8 @@ def action_reset(args, feature_list_path, state_dir):
 
     # Reset runtime fields
     fs["retry_count"] = 0
+    fs["infra_error_count"] = 0
+    fs["last_infra_error_session_id"] = None
     fs["sessions"] = []
     fs["last_session_id"] = None
     fs["resume_from_phase"] = None
@@ -1264,6 +1297,8 @@ def action_clean(args, feature_list_path, state_dir):
     old_retry = fs.get("retry_count", 0)
 
     fs["retry_count"] = 0
+    fs["infra_error_count"] = 0
+    fs["last_infra_error_session_id"] = None
     fs["sessions"] = []
     fs["last_session_id"] = None
     fs["resume_from_phase"] = None
@@ -1423,6 +1458,8 @@ def action_unskip(args, feature_list_path, state_dir):
     for fid in to_reset:
         fs = load_feature_status(state_dir, fid)
         fs["retry_count"] = 0
+        fs["infra_error_count"] = 0
+        fs["last_infra_error_session_id"] = None
         fs["sessions"] = []
         fs["last_session_id"] = None
         fs["resume_from_phase"] = None

package/bundled/dev-pipeline/scripts/update-refactor-status.py

@@ -16,7 +16,7 @@ Usage:
         --refactor-list <path> --state-dir <path> \
         --action <get_next|start|update|status|pause|reset|clean|unskip> \
         [--refactor-id <id>] [--session-status <status>] \
-        [--session-id <id>] [--max-retries <n>]
+        [--session-id <id>] [--max-retries <n>] [--max-infra-retries <n>]
 """
 
 import argparse
@@ -87,7 +87,8 @@ def parse_args():
         help="Session outcome status (required for 'update' action)",
     )
     parser.add_argument("--session-id", default=None, help="Session ID (optional, for 'update' action)")
-    parser.add_argument("--max-retries", type=int, default=3, help="Maximum retry count (default: 3)")
+    parser.add_argument("--max-retries", type=int, default=3, help="Maximum code retry count (default: 3)")
+    parser.add_argument("--max-infra-retries", type=int, default=3, help="Maximum infrastructure retry count (default: 3)")
     parser.add_argument("--project-root", default=None, help="Project root directory. Required for 'clean' action.")
     return parser.parse_args()
 
@@ -103,6 +104,9 @@ def _default_status(refactor_id):
         "refactor_id": refactor_id,
         "retry_count": 0,
         "max_retries": 3,
+        "max_infra_retries": 3,
+        "infra_error_count": 0,
+        "last_infra_error_session_id": None,
         "sessions": [],
         "last_session_id": None,
         "resume_from_phase": None,
@@ -268,6 +272,7 @@ def action_get_next(refactor_list_data, state_dir):
         "priority": chosen.get("priority", "medium"),
         "complexity": chosen.get("complexity", "medium"),
         "retry_count": chosen_status_data.get("retry_count", 0),
+        "infra_error_count": chosen_status_data.get("infra_error_count", 0),
         "resume_from_phase": chosen_status_data.get("resume_from_phase", None),
     }
     print(json.dumps(result, indent=2, ensure_ascii=False))
@@ -282,6 +287,7 @@ def action_update(args, refactor_list_path, state_dir):
     session_status = args.session_status
     session_id = args.session_id
     max_retries = args.max_retries
+    max_infra_retries = args.max_infra_retries
 
     if not refactor_id:
         error_out("--refactor-id is required for 'update' action")
@@ -296,6 +302,8 @@ def action_update(args, refactor_list_path, state_dir):
     new_status = get_refactor_status_from_list(refactor_list_path, refactor_id)
 
     if session_status == "success":
+        rs["infra_error_count"] = 0
+        rs["last_infra_error_session_id"] = None
         new_status = "completed"
         rs["resume_from_phase"] = None
         err = update_refactor_in_list(refactor_list_path, refactor_id, "completed")
@@ -320,11 +328,25 @@ def action_update(args, refactor_list_path, state_dir):
             error_out("Failed to update .prizmkit/plans/refactor-list.json: {}".format(err))
             return
     elif session_status == "infra_error":
-        new_status = "pending"
-        rs["infra_error_count"] = rs.get("infra_error_count", 0) + 1
+        # AI CLI/provider outage, auth failure, gateway error, etc.
+        # Infra failures do not consume the code retry budget, but they still
+        # need their own bounded budget so a flaky provider cannot loop forever.
+        infra_error_count = rs.get("infra_error_count", 0) + 1
+        rs["infra_error_count"] = infra_error_count
         rs["last_infra_error_session_id"] = session_id
+        rs["max_infra_retries"] = max_infra_retries
+        rs["degraded_reason"] = "infra_error"
+        if session_id:
+            rs["last_session_id"] = session_id
         rs["resume_from_phase"] = None
 
+        if infra_error_count >= max_infra_retries:
+            new_status = "failed"
+            if session_id:
+                rs["last_failed_session_id"] = session_id
+        else:
+            new_status = "pending"
+
         err = update_refactor_in_list(refactor_list_path, refactor_id, new_status)
         if err:
             error_out("Failed to update .prizmkit/plans/refactor-list.json: {}".format(err))
@@ -379,6 +401,8 @@ def action_update(args, refactor_list_path, state_dir):
         "session_status": session_status,
         "new_status": new_status,
         "retry_count": rs["retry_count"],
+        "infra_error_count": rs.get("infra_error_count", 0),
+        "max_infra_retries": max_infra_retries,
         "resume_from_phase": rs.get("resume_from_phase"),
         "updated_at": rs["updated_at"],
     }
@@ -728,6 +752,8 @@ def action_reset(args, refactor_list_path, state_dir):
     old_retry = rs.get("retry_count", 0)
 
     rs["retry_count"] = 0
+    rs["infra_error_count"] = 0
+    rs["last_infra_error_session_id"] = None
     rs["sessions"] = []
     rs["last_session_id"] = None
     rs["resume_from_phase"] = None
@@ -799,6 +825,8 @@ def action_clean(args, refactor_list_path, state_dir):
     old_retry = rs.get("retry_count", 0)
 
     rs["retry_count"] = 0
+    rs["infra_error_count"] = 0
+    rs["last_infra_error_session_id"] = None
     rs["sessions"] = []
     rs["last_session_id"] = None
     rs["resume_from_phase"] = None
@@ -887,6 +915,8 @@ def action_start(args, refactor_list_path, state_dir):
         "refactor_id": refactor_id,
         "old_status": old_status,
         "new_status": "in_progress",
+        "retry_count": rs.get("retry_count", 0),
+        "infra_error_count": rs.get("infra_error_count", 0),
         "updated_at": rs["updated_at"],
     }
     print(json.dumps(result, indent=2, ensure_ascii=False))
@@ -1025,6 +1055,8 @@ def action_unskip(args, refactor_list_path, state_dir):
     for rid in to_reset:
         rs = load_refactor_status(state_dir, rid)
         rs["retry_count"] = 0
+        rs["infra_error_count"] = 0
+        rs["last_infra_error_session_id"] = None
         rs["sessions"] = []
         rs["last_session_id"] = None
         rs["resume_from_phase"] = None

package/bundled/dev-pipeline/tests/test_auto_skip.py

@@ -303,7 +303,7 @@ def _run_get_next(fl_path, state_dir):
     return result.stdout.strip()
 
 
-def _run_update(fl_path, state_dir, feature_id, session_status, session_id="session-1", max_retries=3):
+def _run_update(fl_path, state_dir, feature_id, session_status, session_id="session-1", max_retries=3, max_infra_retries=3):
     cmd = [
         "python3", _SCRIPT,
         "--feature-list", fl_path,
@@ -312,6 +312,7 @@ def _run_update(fl_path, state_dir, feature_id, session_status, session_id="sess
         "--session-status", session_status,
         "--session-id", session_id,
         "--max-retries", str(max_retries),
+        "--max-infra-retries", str(max_infra_retries),
         "--action", "update",
     ]
     result = subprocess.run(cmd, capture_output=True, text=True)
@@ -320,7 +321,7 @@ def _run_update(fl_path, state_dir, feature_id, session_status, session_id="sess
 
 
 class TestInfraErrorUpdate:
-    def test_infra_error_keeps_pending_without_consuming_retry(self, tmp_path):
+    def test_infra_error_below_budget_keeps_pending_without_consuming_retry(self, tmp_path):
         features = [_make_feature("F-001", "Root", status="in_progress")]
         fl_path = _write_fl(tmp_path, features)
         state_dir = _init_state(tmp_path, ["F-001"])
@@ -329,10 +330,20 @@ class TestInfraErrorUpdate:
         fs["retry_count"] = 2
         write_json_file(status_path, fs)
 
-        result = _run_update(fl_path, state_dir, "F-001", "infra_error", "session-infra", max_retries=3)
+        result = _run_update(
+            fl_path,
+            state_dir,
+            "F-001",
+            "infra_error",
+            "session-infra",
+            max_retries=3,
+            max_infra_retries=3,
+        )
 
         assert result["new_status"] == "pending"
         assert result["retry_count"] == 2
+        assert result["infra_error_count"] == 1
+        assert result["max_infra_retries"] == 3
         assert result["restart_policy"] == "infra_retry"
         assert _read_statuses(fl_path)["F-001"] == "pending"
 
@@ -340,7 +351,85 @@ class TestInfraErrorUpdate:
         assert fs["retry_count"] == 2
         assert fs["infra_error_count"] == 1
         assert fs["last_infra_error_session_id"] == "session-infra"
+        assert fs["max_infra_retries"] == 3
 
+    def test_infra_error_at_budget_marks_failed_without_consuming_code_retry(self, tmp_path):
+        features = [_make_feature("F-001", "Root", status="in_progress")]
+        fl_path = _write_fl(tmp_path, features)
+        state_dir = _init_state(tmp_path, ["F-001"])
+        status_path = os.path.join(state_dir, "features", "F-001", "status.json")
+        fs = load_feature_status(state_dir, "F-001")
+        fs["retry_count"] = 1
+        fs["infra_error_count"] = 2
+        write_json_file(status_path, fs)
+
+        result = _run_update(
+            fl_path,
+            state_dir,
+            "F-001",
+            "infra_error",
+            "session-infra-final",
+            max_retries=3,
+            max_infra_retries=3,
+        )
+
+        assert result["new_status"] == "failed"
+        assert result["retry_count"] == 1
+        assert result["infra_error_count"] == 3
+        assert result["restart_policy"] == "infra_retry"
+        assert _read_statuses(fl_path)["F-001"] == "failed"
+
+        fs = load_feature_status(state_dir, "F-001")
+        assert fs["retry_count"] == 1
+        assert fs["infra_error_count"] == 3
+        assert fs["last_infra_error_session_id"] == "session-infra-final"
+        assert fs["last_failed_session_id"] == "session-infra-final"
+
+    def test_reset_clears_infra_retry_counters(self, tmp_path):
+        features = [_make_feature("F-001", "Root", status="failed")]
+        fl_path = _write_fl(tmp_path, features)
+        state_dir = _init_state(tmp_path, ["F-001"])
+        status_path = os.path.join(state_dir, "features", "F-001", "status.json")
+        fs = load_feature_status(state_dir, "F-001")
+        fs["retry_count"] = 3
+        fs["infra_error_count"] = 3
+        fs["last_infra_error_session_id"] = "session-infra-final"
+        write_json_file(status_path, fs)
+
+        cmd = [
+            "python3", _SCRIPT,
+            "--feature-list", fl_path,
+            "--state-dir", state_dir,
+            "--feature-id", "F-001",
+            "--action", "reset",
+        ]
+        result = subprocess.run(cmd, capture_output=True, text=True)
+        assert result.returncode == 0, result.stderr
+
+        fs = load_feature_status(state_dir, "F-001")
+        assert fs["retry_count"] == 0
+        assert fs["infra_error_count"] == 0
+        assert fs["last_infra_error_session_id"] is None
+        assert _read_statuses(fl_path)["F-001"] == "pending"
+
+    def test_unskip_clears_infra_retry_counters(self, tmp_path):
+        features = [_make_feature("F-001", "Root", status="failed")]
+        fl_path = _write_fl(tmp_path, features)
+        state_dir = _init_state(tmp_path, ["F-001"])
+        status_path = os.path.join(state_dir, "features", "F-001", "status.json")
+        fs = load_feature_status(state_dir, "F-001")
+        fs["retry_count"] = 3
+        fs["infra_error_count"] = 3
+        fs["last_infra_error_session_id"] = "session-infra-final"
+        write_json_file(status_path, fs)
+
+        _run_unskip(fl_path, state_dir, "F-001")
+
+        fs = load_feature_status(state_dir, "F-001")
+        assert fs["retry_count"] == 0
+        assert fs["infra_error_count"] == 0
+        assert fs["last_infra_error_session_id"] is None
+        assert _read_statuses(fl_path)["F-001"] == "pending"
 
 class TestUnskipByFeatureId:
     """Unskip with --feature-id targets a specific failed feature + downstream."""

package/bundled/dev-pipeline-windows/lib/branch.ps1

@@ -1,6 +1,20 @@
 Set-StrictMode -Version 2.0
 $ErrorActionPreference = 'Stop'
 
+function Get-PrizmGitStatusSafe {
+  param([string]$ProjectRoot)
+  $hiddenToolWorktreeExcludes = Get-PrizmHiddenToolWorktreeExcludes
+  $lines = & git -C $ProjectRoot status --porcelain -- . @hiddenToolWorktreeExcludes 2>$null
+  return @($lines)
+}
+
+function Add-PrizmGitAllSafe {
+  param([string]$ProjectRoot)
+  $hiddenToolWorktreeExcludes = Get-PrizmHiddenToolWorktreeExcludes
+  & git -C $ProjectRoot add -A -- . @hiddenToolWorktreeExcludes *> $null
+  return $LASTEXITCODE -eq 0
+}
+
 function Get-PrizmCurrentBranch {
   param([string]$ProjectRoot)
   $branch = & git -C $ProjectRoot rev-parse --abbrev-ref HEAD 2>$null
@@ -76,13 +90,12 @@ function Save-PrizmBranchWip {
   $currentBranch = Get-PrizmCurrentBranch $ProjectRoot
   if ($currentBranch -ne $DevBranch) { return $true }
 
-  $changes = & git -C $ProjectRoot status --porcelain 2>$null
+  $changes = Get-PrizmGitStatusSafe $ProjectRoot
   if ([string]::IsNullOrWhiteSpace(($changes -join "`n"))) { return $true }
 
   Write-PrizmWarn "Saving uncommitted work-in-progress on branch: $DevBranch"
-  & git -C $ProjectRoot add -A *> $null
-  if ($LASTEXITCODE -ne 0) {
-    Write-PrizmWarn "git add -A failed - uncommitted work may remain on $DevBranch"
+  if (-not (Add-PrizmGitAllSafe $ProjectRoot)) {
+    Write-PrizmWarn "git add failed - uncommitted work may remain on $DevBranch"
     return $true
   }
 

package/bundled/dev-pipeline-windows/lib/common.ps1

@@ -51,6 +51,15 @@ function Write-PrizmWarn { param([string]$Message) Write-Host "[WARN]    $(Get-D
 function Write-PrizmError { param([string]$Message) Write-Host "[ERROR]   $(Get-Date -Format 'yyyy-MM-dd HH:mm:ss') $Message" -ForegroundColor Red }
 function Write-PrizmSuccess { param([string]$Message) Write-Host "[SUCCESS] $(Get-Date -Format 'yyyy-MM-dd HH:mm:ss') $Message" -ForegroundColor Green }
 
+function Get-PrizmHiddenToolWorktreeExcludes {
+  return @(
+    ':(top,exclude,glob).*/worktree',
+    ':(top,exclude,glob).*/worktree/**',
+    ':(top,exclude,glob).*/worktrees',
+    ':(top,exclude,glob).*/worktrees/**'
+  )
+}
+
 function Initialize-PrizmPaths {
   param([string]$ScriptRoot)
   $pipelineDir = Resolve-Path $ScriptRoot