prizmkit 1.1.192 → 1.1.193

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (56) hide show
  1. package/bundled/VERSION.json +3 -3
  2. package/bundled/dev-pipeline/README.md +23 -34
  3. package/bundled/dev-pipeline/assets/feature-list-example.json +0 -1
  4. package/bundled/dev-pipeline/assets/skill-subagent-integration.md +3 -0
  5. package/bundled/dev-pipeline/prizmkit_runtime/config.py +24 -2
  6. package/bundled/dev-pipeline/prizmkit_runtime/daemon.py +2 -0
  7. package/bundled/dev-pipeline/prizmkit_runtime/gitops.py +89 -40
  8. package/bundled/dev-pipeline/prizmkit_runtime/heartbeat.py +9 -1
  9. package/bundled/dev-pipeline/prizmkit_runtime/interoperability.py +0 -22
  10. package/bundled/dev-pipeline/prizmkit_runtime/platform_detection.py +11 -7
  11. package/bundled/dev-pipeline/prizmkit_runtime/processes.py +26 -20
  12. package/bundled/dev-pipeline/prizmkit_runtime/reset.py +37 -4
  13. package/bundled/dev-pipeline/prizmkit_runtime/reset_preserve.py +20 -41
  14. package/bundled/dev-pipeline/prizmkit_runtime/runner_bookkeeping.py +74 -36
  15. package/bundled/dev-pipeline/prizmkit_runtime/runner_models.py +0 -66
  16. package/bundled/dev-pipeline/prizmkit_runtime/runners.py +271 -52
  17. package/bundled/dev-pipeline/prizmkit_runtime/runtime_commit.py +51 -13
  18. package/bundled/dev-pipeline/prizmkit_runtime/sessions.py +43 -15
  19. package/bundled/dev-pipeline/prizmkit_runtime/stream_progress.py +27 -24
  20. package/bundled/dev-pipeline/prizmkit_runtime/task_checkout.py +0 -1
  21. package/bundled/dev-pipeline/scripts/check-session-status.py +13 -1
  22. package/bundled/dev-pipeline/scripts/continuation.py +2 -3
  23. package/bundled/dev-pipeline/scripts/detect-stuck.py +19 -5
  24. package/bundled/dev-pipeline/scripts/generate-bootstrap-prompt.py +6 -14
  25. package/bundled/dev-pipeline/scripts/generate-bugfix-prompt.py +13 -8
  26. package/bundled/dev-pipeline/scripts/init-change-artifact.py +17 -2
  27. package/bundled/dev-pipeline/scripts/prompt_framework.py +39 -21
  28. package/bundled/dev-pipeline/scripts/update-checkpoint.py +27 -0
  29. package/bundled/dev-pipeline/templates/bootstrap-tier3.md +1 -1
  30. package/bundled/dev-pipeline/templates/bugfix-bootstrap-prompt.md +6 -6
  31. package/bundled/dev-pipeline/templates/sections/bugfix-phase-test.md +1 -1
  32. package/bundled/dev-pipeline/templates/sections/phase-prizmkit-test.md +1 -1
  33. package/bundled/skills/_metadata.json +1 -1
  34. package/package.json +1 -1
  35. package/bundled/dev-pipeline/tests/_helpers.py +0 -418
  36. package/bundled/dev-pipeline/tests/conftest.py +0 -73
  37. package/bundled/dev-pipeline/tests/live_ai_cli_log_smoke.py +0 -208
  38. package/bundled/dev-pipeline/tests/test_auto_skip.py +0 -1806
  39. package/bundled/dev-pipeline/tests/test_checkpoint_state.py +0 -1439
  40. package/bundled/dev-pipeline/tests/test_daemon_and_scripts.py +0 -257
  41. package/bundled/dev-pipeline/tests/test_e2e_pipeline_run.py +0 -500
  42. package/bundled/dev-pipeline/tests/test_generate_bootstrap_prompt.py +0 -1654
  43. package/bundled/dev-pipeline/tests/test_generate_bugfix_prompt.py +0 -629
  44. package/bundled/dev-pipeline/tests/test_python_runner_parity.py +0 -2973
  45. package/bundled/dev-pipeline/tests/test_recovery_workflow.py +0 -321
  46. package/bundled/dev-pipeline/tests/test_repository_bootstrap.py +0 -251
  47. package/bundled/dev-pipeline/tests/test_reset_modes.py +0 -1196
  48. package/bundled/dev-pipeline/tests/test_reset_preserve.py +0 -872
  49. package/bundled/dev-pipeline/tests/test_runner_bookkeeping.py +0 -766
  50. package/bundled/dev-pipeline/tests/test_runner_lifecycle.py +0 -1257
  51. package/bundled/dev-pipeline/tests/test_runtime_commit.py +0 -1113
  52. package/bundled/dev-pipeline/tests/test_runtime_helper.py +0 -380
  53. package/bundled/dev-pipeline/tests/test_session_launch.py +0 -1101
  54. package/bundled/dev-pipeline/tests/test_stream_progress.py +0 -585
  55. package/bundled/dev-pipeline/tests/test_unified_cli.py +0 -3625
  56. package/bundled/dev-pipeline/tests/test_utils.py +0 -170
@@ -184,11 +184,14 @@ def detect_stream_json_support(
184
184
  if len(raw_command) >= 2 and raw_command[0] == raw_command[-1] and raw_command[0] in "'\""
185
185
  else raw_command
186
186
  )
187
- command_token = (
188
- unquoted
189
- if len(unquoted) >= 3 and unquoted[1:3] in {":\\", ":/"} and unquoted.lower().endswith((".exe", ".cmd", ".bat"))
190
- else raw_command.split()[0].strip("'\"")
191
- )
187
+ if len(unquoted) >= 3 and unquoted[1:3] in {":\\", ":/"} and unquoted.lower().endswith((".exe", ".cmd", ".bat")):
188
+ command_token = unquoted
189
+ else:
190
+ try:
191
+ tokens = shlex.split(raw_command)
192
+ command_token = tokens[0] if tokens else raw_command
193
+ except ValueError:
194
+ command_token = raw_command.split()[0].strip("'\"")
192
195
  command_name = command_token.replace("\\", "/").rsplit("/", 1)[-1].lower()
193
196
  for suffix in (".exe", ".cmd", ".bat"):
194
197
  if command_name.endswith(suffix):
@@ -1293,8 +1296,10 @@ class AISessionLauncher:
1293
1296
  display_heartbeat_state = None
1294
1297
  handle: ProcessHandle | None = None
1295
1298
  reader: threading.Thread | None = None
1299
+ stdin_writer: threading.Thread | None = None
1296
1300
  heartbeat: HeartbeatMonitor | None = None
1297
1301
  interrupted = False
1302
+ launcher_aborted = False
1298
1303
  reader_result_queue: queue.Queue[dict[str, object]] = queue.Queue(maxsize=1)
1299
1304
 
1300
1305
  try:
@@ -1314,13 +1319,6 @@ class AISessionLauncher:
1314
1319
  daemon=True,
1315
1320
  )
1316
1321
  reader.start()
1317
- if handle.popen.stdin is not None:
1318
- try:
1319
- if command.stdin_text is not None:
1320
- handle.popen.stdin.write(command.stdin_text)
1321
- handle.popen.stdin.close()
1322
- except (BrokenPipeError, OSError):
1323
- pass
1324
1322
 
1325
1323
  heartbeat_path = self.config.heartbeat_path
1326
1324
  if heartbeat_path is None and kill_threshold_seconds > 0:
@@ -1341,6 +1339,25 @@ class AISessionLauncher:
1341
1339
  )
1342
1340
  heartbeat.start()
1343
1341
 
1342
+ if handle.popen.stdin is not None:
1343
+ stdin_pipe = handle.popen.stdin
1344
+ stdin_text = command.stdin_text
1345
+
1346
+ def _write_stdin() -> None:
1347
+ try:
1348
+ if stdin_text is not None:
1349
+ stdin_pipe.write(stdin_text)
1350
+ except (BrokenPipeError, OSError):
1351
+ pass
1352
+ finally:
1353
+ try:
1354
+ stdin_pipe.close()
1355
+ except (BrokenPipeError, OSError):
1356
+ pass
1357
+
1358
+ stdin_writer = threading.Thread(target=_write_stdin, name="prizmkit-stdin", daemon=True)
1359
+ stdin_writer.start()
1360
+
1344
1361
  while True:
1345
1362
  timeout = 0.2
1346
1363
  if self.config.total_timeout_seconds is not None:
@@ -1421,8 +1438,19 @@ class AISessionLauncher:
1421
1438
  )
1422
1439
  _emit_ai_interruption_termination(termination)
1423
1440
  raise
1441
+ except Exception:
1442
+ launcher_aborted = True
1443
+ if handle is not None:
1444
+ terminate_process_tree(
1445
+ handle,
1446
+ "launcher_error",
1447
+ grace_seconds=self.config.stale_kill_grace_seconds,
1448
+ )
1449
+ raise
1424
1450
  finally:
1425
- if reader is not None and not interrupted:
1451
+ if stdin_writer is not None:
1452
+ stdin_writer.join(timeout=1.0)
1453
+ if reader is not None and not interrupted and not launcher_aborted:
1426
1454
  reader.join(timeout=1.5)
1427
1455
  if (
1428
1456
  reader.is_alive()
@@ -1433,9 +1461,9 @@ class AISessionLauncher:
1433
1461
  terminate_process_tree(handle, "lingering_child_pipes", grace_seconds=0.5)
1434
1462
  reader.join(timeout=3.5)
1435
1463
  heartbeat_result = heartbeat.stop() if heartbeat is not None else None
1436
- if reader is not None and interrupted:
1464
+ if reader is not None and (interrupted or launcher_aborted):
1437
1465
  reader.join(timeout=1.0)
1438
- if not interrupted and heartbeat_result and heartbeat_result.killed:
1466
+ if not interrupted and not launcher_aborted and heartbeat_result and heartbeat_result.killed:
1439
1467
  termination_reason = heartbeat_result.reason
1440
1468
  stale_marker = heartbeat_result.marker_path
1441
1469
  if interrupted:
@@ -2135,11 +2135,20 @@ class ProgressTracker:
2135
2135
  text_parts = []
2136
2136
  terminal_parts = []
2137
2137
 
2138
- for key in ("fatal_error_code", "api_error_code", "error_code", "stop_reason"):
2138
+ for key in ("fatal_error_code", "api_error_code", "error_code"):
2139
2139
  value = event.get(key)
2140
2140
  if isinstance(value, (str, int, float)) and str(value).strip():
2141
2141
  code_parts.append(str(value))
2142
2142
 
2143
+ raw_stop_reason = event.get("stop_reason")
2144
+ stop_reason_code = ""
2145
+ if isinstance(raw_stop_reason, (str, int, float)) and str(raw_stop_reason).strip():
2146
+ stop_reason_code = detect_api_error_code(
2147
+ str(raw_stop_reason).strip(),
2148
+ require_error_context=False,
2149
+ structured_code=True,
2150
+ )
2151
+
2143
2152
  for key in ("terminal_result_text", "message", "result"):
2144
2153
  value = event.get(key)
2145
2154
  if isinstance(value, (str, int, float)) and str(value).strip():
@@ -2159,24 +2168,29 @@ class ProgressTracker:
2159
2168
 
2160
2169
  combined_code_text = " ".join(code_parts)
2161
2170
  combined_text = " ".join(code_parts + text_parts)
2162
- if not combined_text:
2171
+ if not combined_text and not stop_reason_code:
2163
2172
  return
2164
2173
 
2165
2174
  error_like = (
2166
2175
  bool(code_parts)
2176
+ or bool(stop_reason_code)
2167
2177
  or event.get("api_error_status") not in (None, "")
2168
2178
  or bool(event.get("last_result_is_error"))
2169
2179
  or bool(event.get("is_error"))
2170
2180
  or _has_error_context(combined_text)
2171
2181
  )
2172
- normalized_code = detect_api_error_code(
2173
- combined_code_text,
2174
- require_error_context=False,
2175
- structured_code=True,
2176
- ) or detect_api_error_code(
2177
- combined_text,
2178
- require_error_context=not error_like,
2179
- structured_code=bool(code_parts),
2182
+ normalized_code = (
2183
+ detect_api_error_code(
2184
+ combined_code_text,
2185
+ require_error_context=False,
2186
+ structured_code=True,
2187
+ )
2188
+ or stop_reason_code
2189
+ or detect_api_error_code(
2190
+ combined_text,
2191
+ require_error_context=not error_like,
2192
+ structured_code=bool(code_parts),
2193
+ )
2180
2194
  )
2181
2195
  if not normalized_code:
2182
2196
  return
@@ -2255,6 +2269,9 @@ class ProgressTracker:
2255
2269
  """Record a Claude Code terminal result event."""
2256
2270
  terminal_text = str(text or "")
2257
2271
  self.last_result_is_error = bool(is_error)
2272
+ normalized_stop_reason = str(stop_reason or "").strip()
2273
+ if normalized_stop_reason:
2274
+ self.stop_reason = normalized_stop_reason
2258
2275
  if api_error_status not in (None, ""):
2259
2276
  try:
2260
2277
  self.api_error_status = int(api_error_status)
@@ -2498,13 +2515,6 @@ class ProgressTracker:
2498
2515
  return Path(claude_home).expanduser() / "projects"
2499
2516
  return Path.home() / ".claude" / "projects"
2500
2517
 
2501
- def _claude_project_key(self):
2502
- """Encode cwd the same way Claude Code stores project transcript dirs."""
2503
- cwd = self.claude_cwd
2504
- if not cwd:
2505
- return ""
2506
- return cwd.replace("\\", "-").replace("/", "-").replace(":", "")
2507
-
2508
2518
  def _claude_candidate_project_keys(self, cwd: str) -> list[str]:
2509
2519
  if not cwd:
2510
2520
  return []
@@ -2569,13 +2579,6 @@ class ProgressTracker:
2569
2579
  return Path(cb_home).expanduser() / "projects"
2570
2580
  return Path.home() / ".codebuddy" / "projects"
2571
2581
 
2572
- def _cb_project_key(self):
2573
- """Encode cwd the same way CodeBuddy stores project subdirs."""
2574
- cwd = self.cb_cwd
2575
- if not cwd:
2576
- return ""
2577
- return cwd.replace("\\", "-").replace("/", "-").replace(":", "").lstrip("-")
2578
-
2579
2582
  def _cb_candidate_project_keys(self, cwd: str) -> list[str]:
2580
2583
  if not cwd:
2581
2584
  return []
@@ -3,7 +3,6 @@
3
3
  from __future__ import annotations
4
4
 
5
5
  import json
6
- import os
7
6
  from contextlib import contextmanager
8
7
  from dataclasses import asdict, dataclass
9
8
  from pathlib import Path
@@ -175,7 +175,19 @@ def determine_status(data):
175
175
 
176
176
  if status == "success":
177
177
  return "success"
178
- if status == "context_overflow" or has_context_overflow_signal(data):
178
+ if status == "context_overflow":
179
+ return "context_overflow"
180
+ explicit_terminal_statuses = {
181
+ "stalled_context_continuation",
182
+ "commit_missing",
183
+ "docs_missing",
184
+ "merge_conflict",
185
+ "workflow_blocked",
186
+ "workflow_skipped",
187
+ }
188
+ if status in explicit_terminal_statuses:
189
+ return status
190
+ if has_context_overflow_signal(data):
179
191
  return "context_overflow"
180
192
  elif status == "partial":
181
193
  can_resume = data.get("can_resume", False)
@@ -8,6 +8,7 @@ _PIPELINE_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
8
8
  if _PIPELINE_ROOT not in sys.path:
9
9
  sys.path.insert(0, _PIPELINE_ROOT)
10
10
 
11
+ from prizmkit_runtime.atomic_io import write_text_atomic # noqa: E402
11
12
  from prizmkit_runtime.checkpoint_state import checkpoint_cursor as shared_checkpoint_cursor # noqa: E402
12
13
 
13
14
  SUPPORTED_CONTINUATION_MODES = {"context_overflow", "ai_error"}
@@ -376,9 +377,7 @@ def append_continuation_handoff(rendered_prompt, args, project_root, task_type,
376
377
  summary_text = _build_summary(context)
377
378
  for summary_path in paths:
378
379
  abs_summary = os.path.join(project_root, summary_path)
379
- os.makedirs(os.path.dirname(abs_summary), exist_ok=True)
380
- with open(abs_summary, "w", encoding="utf-8") as f:
381
- f.write(summary_text)
380
+ write_text_atomic(abs_summary, summary_text)
382
381
 
383
382
  addendum = _build_addendum(context)
384
383
  return rendered_prompt.rstrip() + "\n\n" + addendum
@@ -252,11 +252,16 @@ def check_stale_heartbeat(item_id, item_status, state_dir, items_subdir, stale_t
252
252
  if not session_id:
253
253
  return None
254
254
 
255
- # Check heartbeat file
255
+ # Check heartbeat file (canonical logs/heartbeat.json first, legacy fallback)
256
256
  heartbeat_path = os.path.join(
257
+ state_dir, items_subdir, item_id, "sessions", session_id, "logs", "heartbeat.json"
258
+ )
259
+ legacy_heartbeat_path = os.path.join(
257
260
  state_dir, items_subdir, item_id, "sessions", session_id, "heartbeat.json"
258
261
  )
259
262
  heartbeat = load_json(heartbeat_path)
263
+ if heartbeat is None and os.path.exists(legacy_heartbeat_path):
264
+ heartbeat = load_json(legacy_heartbeat_path)
260
265
 
261
266
  if heartbeat is None:
262
267
  return {
@@ -268,16 +273,22 @@ def check_stale_heartbeat(item_id, item_status, state_dir, items_subdir, stale_t
268
273
  }
269
274
 
270
275
  # Parse heartbeat timestamp and check staleness
276
+ epoch_val = heartbeat.get("last_seen_epoch")
271
277
  ts_str = (
272
278
  heartbeat.get("timestamp")
273
279
  or heartbeat.get("last_heartbeat_at")
274
280
  or heartbeat.get("updated_at")
275
281
  )
276
- heartbeat_time = parse_iso_timestamp(ts_str)
282
+ if isinstance(epoch_val, (int, float)) and not isinstance(epoch_val, bool):
283
+ heartbeat_time = datetime.fromtimestamp(float(epoch_val), tz=timezone.utc)
284
+ else:
285
+ heartbeat_time = parse_iso_timestamp(ts_str)
277
286
  if heartbeat_time is None:
278
287
  return {
279
288
  "reason": "stale_heartbeat",
280
- "details": "Heartbeat has invalid or unparseable timestamp: {}".format(ts_str),
289
+ "details": "Heartbeat has invalid or unparseable timestamp: {}".format(
290
+ epoch_val if epoch_val is not None else ts_str
291
+ ),
281
292
  "suggestion": "Check the agent session - heartbeat timestamp is malformed",
282
293
  }
283
294
 
@@ -398,6 +409,9 @@ def check_item(item_id, state_dir, items_subdir, items_key, task_list_data, max_
398
409
  task_list_status = item.get("status", "pending")
399
410
  break
400
411
 
412
+ if task_list_status in {"completed", "failed", "skipped", "auto_skipped", "needs_info", "split"}:
413
+ return []
414
+
401
415
  reports = []
402
416
 
403
417
  # Check 1: Max retries exceeded
@@ -432,8 +446,8 @@ def resolve_pipeline_type(args):
432
446
  # Explicit --pipeline-type takes precedence
433
447
  if args.pipeline_type:
434
448
  ptype = args.pipeline_type
435
- item_id = args.item_id
436
- task_list = args.task_list
449
+ item_id = args.item_id or (args.feature_id if ptype == "feature" else args.bug_id)
450
+ task_list = args.task_list or (args.feature_list if ptype == "feature" else args.bug_list)
437
451
  # Legacy / shorthand: --feature-id, --bug-id
438
452
  elif args.feature_id:
439
453
  ptype = "feature"
@@ -24,7 +24,7 @@ import os
24
24
  import re
25
25
  import sys
26
26
 
27
- from utils import enrich_global_context, helper_replacements, load_json_file, resolve_prompt_platform, setup_logging
27
+ from utils import enrich_global_context, helper_replacements, load_json_file, resolve_prompt_platform, setup_logging, write_json_file
28
28
  from continuation import add_continuation_args, append_continuation_handoff, is_continuation
29
29
  from prompt_framework import (
30
30
  artifact_exists,
@@ -268,19 +268,15 @@ def extract_baseline_failures(test_cmd, project_root):
268
268
 
269
269
  try:
270
270
  import subprocess
271
- original_cwd = os.getcwd()
272
- os.chdir(project_root)
273
-
274
271
  result = subprocess.run(
275
272
  test_cmd,
276
273
  shell=True,
274
+ cwd=project_root,
277
275
  capture_output=True,
278
276
  text=True,
279
- timeout=120
277
+ timeout=120,
280
278
  )
281
279
 
282
- os.chdir(original_cwd)
283
-
284
280
  output = result.stdout + result.stderr
285
281
  failures = []
286
282
 
@@ -296,11 +292,6 @@ def extract_baseline_failures(test_cmd, project_root):
296
292
 
297
293
  except Exception as e:
298
294
  return f"(error: {str(e)})"
299
- finally:
300
- try:
301
- os.chdir(original_cwd)
302
- except Exception:
303
- pass
304
295
 
305
296
 
306
297
  def format_ac_checklist(acceptance_criteria):
@@ -1273,8 +1264,9 @@ def main():
1273
1264
  )
1274
1265
 
1275
1266
  if not args.no_checkpoint_write:
1276
- with open(checkpoint_path, "w", encoding="utf-8") as f:
1277
- json.dump(checkpoint, f, indent=2, ensure_ascii=False)
1267
+ err = write_json_file(checkpoint_path, checkpoint)
1268
+ if err:
1269
+ emit_failure(err)
1278
1270
  LOGGER.info("Wrote checkpoint to %s", checkpoint_path)
1279
1271
 
1280
1272
  if is_continuation(args):
@@ -27,7 +27,15 @@ from prompt_framework import (
27
27
  validate_rendered,
28
28
  write_output,
29
29
  )
30
- from utils import enrich_global_context, helper_replacements, load_json_file, read_checkpoint_system_section, resolve_prompt_platform, setup_logging
30
+ from utils import (
31
+ enrich_global_context,
32
+ helper_replacements,
33
+ load_json_file,
34
+ read_checkpoint_system_section,
35
+ resolve_prompt_platform,
36
+ setup_logging,
37
+ write_json_file,
38
+ )
31
39
 
32
40
 
33
41
  LOGGER = setup_logging("generate-bugfix-prompt")
@@ -464,11 +472,7 @@ def generate_bugfix_checkpoint(
464
472
 
465
473
  def merge_bugfix_checkpoint_state(existing, fresh, project_root):
466
474
  """Merge existing bugfix checkpoint state into fresh definition."""
467
- merged = merge_checkpoint_state(existing, fresh, project_root, logger=LOGGER)
468
- existing_state = existing.get("bugfix_state")
469
- if isinstance(existing_state, dict):
470
- merged["bugfix_state"] = existing_state
471
- return merged
475
+ return merge_checkpoint_state(existing, fresh, project_root, logger=LOGGER)
472
476
 
473
477
 
474
478
  def emit_failure(message):
@@ -571,8 +575,9 @@ def main():
571
575
  LOGGER.warning("Existing bugfix checkpoint corrupted (%s) — generating fresh", exc)
572
576
 
573
577
  if not args.no_checkpoint_write:
574
- with open(checkpoint_path, "w", encoding="utf-8") as f:
575
- json.dump(checkpoint, f, indent=2, ensure_ascii=False)
578
+ err = write_json_file(checkpoint_path, checkpoint)
579
+ if err:
580
+ emit_failure(err)
576
581
  LOGGER.info("Wrote bugfix checkpoint to %s", checkpoint_path)
577
582
 
578
583
  if is_continuation(args):
@@ -48,6 +48,17 @@ def create_directories(project_root, feature_slug=None):
48
48
  dirs_to_create = []
49
49
 
50
50
  if feature_slug:
51
+ if (
52
+ os.path.isabs(feature_slug)
53
+ or "/" in feature_slug
54
+ or "\\" in feature_slug
55
+ or feature_slug in {".", ".."}
56
+ ):
57
+ raise ValueError(f"Invalid feature_slug '{feature_slug}'")
58
+ specs_root = os.path.realpath(os.path.join(project_root, ".prizmkit", "specs"))
59
+ target_dir = os.path.realpath(os.path.join(specs_root, feature_slug))
60
+ if os.path.commonpath([specs_root, target_dir]) != specs_root or target_dir == specs_root:
61
+ raise ValueError(f"feature_slug escapes .prizmkit/specs: '{feature_slug}'")
51
62
  dirs_to_create.extend([
52
63
  ".prizmkit/specs/{}".format(feature_slug),
53
64
  ])
@@ -71,8 +82,12 @@ def init_prizmkit_config(project_root, feature_id):
71
82
  config_path = os.path.join(project_root, ".prizmkit", "config.json")
72
83
 
73
84
  if os.path.exists(config_path):
74
- with open(config_path, "r", encoding="utf-8") as f:
75
- config = json.load(f)
85
+ try:
86
+ with open(config_path, "r", encoding="utf-8") as f:
87
+ loaded = json.load(f)
88
+ config = loaded if isinstance(loaded, dict) else {}
89
+ except (json.JSONDecodeError, IOError):
90
+ config = {}
76
91
  config["current_feature"] = feature_id
77
92
  config["updated_at"] = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
78
93
  else:
@@ -14,6 +14,7 @@ _PIPELINE_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
14
14
  if _PIPELINE_ROOT not in sys.path:
15
15
  sys.path.insert(0, _PIPELINE_ROOT)
16
16
 
17
+ from prizmkit_runtime.atomic_io import write_text_atomic # noqa: E402
17
18
  from prizmkit_runtime.checkpoint_state import ( # noqa: E402
18
19
  VALID_CHECKPOINT_STATUSES,
19
20
  VALID_REPAIR_SCOPES,
@@ -230,18 +231,11 @@ def validate_rendered(content, required_markers=None, logger=None):
230
231
 
231
232
 
232
233
  def write_output(output_path, content):
233
- """Write rendered prompt content to output_path."""
234
+ """Write rendered prompt content atomically to output_path."""
234
235
  abs_path = os.path.abspath(output_path)
235
- output_dir = os.path.dirname(abs_path)
236
- if output_dir and not os.path.isdir(output_dir):
237
- try:
238
- os.makedirs(output_dir, exist_ok=True)
239
- except OSError as exc:
240
- return "Cannot create output directory: {}".format(str(exc))
241
236
  try:
242
- with open(abs_path, "w", encoding="utf-8") as f:
243
- f.write(content)
244
- except IOError as exc:
237
+ write_text_atomic(abs_path, content)
238
+ except OSError as exc:
245
239
  return "Cannot write output file: {}".format(str(exc))
246
240
  return None
247
241
 
@@ -427,7 +421,8 @@ def merge_checkpoint_state(existing, fresh, project_root, validator=None, logger
427
421
  return None
428
422
 
429
423
  chain_broken = False
430
- for step in fresh.get("steps", []):
424
+ first_break_index = None
425
+ for index, step in enumerate(fresh.get("steps", [])):
431
426
  if chain_broken:
432
427
  step["status"] = "pending"
433
428
  continue
@@ -463,6 +458,7 @@ def merge_checkpoint_state(existing, fresh, project_root, validator=None, logger
463
458
  step["status"] = "skipped"
464
459
  continue
465
460
  chain_broken = True
461
+ first_break_index = index
466
462
  step["status"] = "pending"
467
463
 
468
464
  workflow_type = fresh.get("workflow_type", "")
@@ -471,31 +467,53 @@ def merge_checkpoint_state(existing, fresh, project_root, validator=None, logger
471
467
  "bugfix-pipeline": "bugfix_state",
472
468
  }.get(workflow_type)
473
469
  if state_key:
470
+ existing_state = existing.get(state_key) if isinstance(existing.get(state_key), dict) else {}
474
471
  merged_state = dict(fresh[state_key]) if isinstance(fresh.get(state_key), dict) else {}
475
- if isinstance(existing.get(state_key), dict):
476
- merged_state.update(existing[state_key])
472
+ if existing_state:
473
+ merged_state.update(existing_state)
477
474
  if chain_broken and merged_state:
478
475
  from prizmkit_runtime.checkpoint_state import (
479
476
  SKILL_WORKFLOW_STAGES,
480
477
  TERMINAL_STEP_STATUSES,
481
478
  )
482
479
 
480
+ steps_list = fresh.get("steps", [])
483
481
  first_active = next(
484
- (s for s in fresh.get("steps", []) if s.get("status") not in TERMINAL_STEP_STATUSES),
482
+ (s for s in steps_list if s.get("status") not in TERMINAL_STEP_STATUSES),
485
483
  None,
486
484
  )
487
485
  next_skill = first_active.get("skill") if isinstance(first_active, dict) else None
486
+ repair_target = existing_state.get("next_stage")
487
+ repair_target_index = next(
488
+ (i for i, s in enumerate(steps_list) if s.get("skill") == repair_target),
489
+ None,
490
+ )
491
+ active_routed_repair = (
492
+ existing_state.get("stage_result") in {"REVIEW_NEEDS_FIXES", "TEST_NEEDS_FIXES"}
493
+ and repair_target == "prizmkit-implement"
494
+ and existing_state.get("terminal_status") is None
495
+ and repair_target_index is not None
496
+ and first_break_index is not None
497
+ and first_break_index >= repair_target_index
498
+ )
488
499
  merged_state["terminal_status"] = None
489
500
  merged_state["blocked_reason"] = None
490
- merged_state["repair_scope"] = None
491
- merged_state.pop("stage_result", None)
492
501
  merged_state.pop("runtime_base_head", None)
493
502
  merged_state.pop("runtime_commit_hash", None)
494
- merged_state["status"] = "pending"
495
- merged_state["current_stage"] = next_skill
496
- merged_state["next_stage"] = next_skill
497
- if next_skill:
498
- merged_state["stage"] = SKILL_WORKFLOW_STAGES.get(next_skill, next_skill)
503
+ if active_routed_repair:
504
+ merged_state["next_stage"] = repair_target
505
+ merged_state["resume_from"] = repair_target
506
+ if not merged_state.get("current_stage") and next_skill:
507
+ merged_state["current_stage"] = next_skill
508
+ else:
509
+ merged_state["repair_scope"] = None
510
+ merged_state.pop("stage_result", None)
511
+ merged_state["status"] = "pending"
512
+ merged_state["current_stage"] = next_skill
513
+ merged_state["next_stage"] = next_skill
514
+ merged_state["resume_from"] = next_skill
515
+ if next_skill:
516
+ merged_state["stage"] = SKILL_WORKFLOW_STAGES.get(next_skill, next_skill)
499
517
  if merged_state:
500
518
  fresh[state_key] = merged_state
501
519
 
@@ -447,6 +447,14 @@ def update_checkpoint(
447
447
  "instead of continuing step mutations."
448
448
  ),
449
449
  }
450
+ if l4_state.get("terminal_status") == "WORKFLOW_COMPLETED":
451
+ return {
452
+ "ok": False,
453
+ "error": (
454
+ "Workflow checkpoint is already WORKFLOW_COMPLETED; reset or rewind the task "
455
+ "instead of continuing step mutations."
456
+ ),
457
+ }
450
458
 
451
459
  existing_round = l4_state.get("repair_round", 0)
452
460
  requested_round = semantic_update.get("repair_round")
@@ -480,6 +488,25 @@ def update_checkpoint(
480
488
 
481
489
  stage_result = semantic_update.get("stage_result")
482
490
  project_root = _checkpoint_project_root(checkpoint_path)
491
+ if new_status == "completed" and workflow_type in {"feature-pipeline", "bugfix-pipeline"}:
492
+ required_completed_results = {
493
+ "prizmkit-plan": "PLAN_READY",
494
+ "context-snapshot-and-plan": "PLAN_READY",
495
+ "bug-diagnosis-and-plan": "PLAN_READY",
496
+ "prizmkit-implement": "IMPLEMENTED",
497
+ "prizmkit-retrospective": "RETRO_COMPLETE",
498
+ }
499
+ expected_completed_result = required_completed_results.get(step.get("skill"))
500
+ if expected_completed_result is not None:
501
+ resolved_result = stage_result or step.get("stage_result")
502
+ if resolved_result != expected_completed_result:
503
+ return {
504
+ "ok": False,
505
+ "error": "{} completion requires {}".format(
506
+ step.get("skill"), expected_completed_result
507
+ ),
508
+ }
509
+ stage_result = resolved_result
483
510
  if new_status == "completed" and step.get("skill") == "prizmkit-code-review":
484
511
  stage_result = stage_result or step.get("stage_result")
485
512
  if stage_result != "REVIEW_PASS":
@@ -406,7 +406,7 @@ After configured Browser/Manual verification, invoke the Feature test gate:
406
406
  /prizmkit-test scope=this-change artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/
407
407
  ```
408
408
 
409
- Consume `.prizmkit/specs/{{FEATURE_SLUG}}/test-report.md` and `test-result.json`. They must agree. `TEST_PASS` completes the gate. For `TEST_NEEDS_FIXES`, classify the concrete correction into an allowed `repair_scope`; safe in-budget repair re-enters Implement through the helper and requires fresh Review, configured verification, and Test, while unknown/unsafe scope or exhausted budget blocks. `TEST_BLOCKED` preserves the blocker and becomes `WORKFLOW_BLOCKED`. Do not require an evidence package, manifest, hashes, attestation, or test-internal checkpoint, and do not reinterpret a testing result as an AI CLI crash.
409
+ Consume `.prizmkit/specs/{{FEATURE_SLUG}}/test-report.md` and `test-result.json`. They must agree. `TEST_PASS` with consistent `production_changed=false` completes the gate; `TEST_PASS` with consistent `production_changed=true` keeps `prizmkit-test` incomplete and re-enters `prizmkit-code-review` in delta mode within the shared 3-round outer repair budget (`repair_round_limit` records `WORKFLOW_BLOCKED`). For `TEST_NEEDS_FIXES`, classify the concrete correction into an allowed `repair_scope`; safe in-budget repair re-enters Implement through the helper and requires fresh Review, configured verification, and Test, while unknown/unsafe scope or exhausted budget blocks. `TEST_BLOCKED` preserves the blocker and becomes `WORKFLOW_BLOCKED`. Do not require an evidence package, manifest, hashes, attestation, or test-internal checkpoint, and do not reinterpret a testing result as an AI CLI crash.
410
410
 
411
411
  Record only caller-owned `stage`, `status`, `stage_result`, `repair_scope`, `repair_round`, `next_stage`, and `resume_from` in recovery metadata. Keep testing reports/results separate from `workflow-checkpoint.json`. The skill owns its internal ten-round Main-Agent review, five-response optional independent review, and three-round execution-failure repair budgets.
412
412
 
@@ -207,12 +207,12 @@ After valid code-review PASS and before Test, execute the configured Browser Ver
207
207
 
208
208
  {{END_IF_BROWSER_INTERACTION}}
209
209
  {{IF_VERIFICATION_MANUAL_OR_HYBRID}}
210
- ### Phase 4: Manual / Hybrid Verification
210
+ ### Phase 4b: Manual / Hybrid Verification
211
211
 
212
212
  Complete the declared UAT checklist automatically with available browser, CLI, API, fixture, and test evidence. Record the evidence before Test. If required evidence cannot be established, write `failure-log.md` and stop truthfully.
213
213
 
214
214
  {{END_IF_VERIFICATION_MANUAL_OR_HYBRID}}
215
- ### Phase 4: Post-Review Test Gate — PrizmKit Test
215
+ ### Phase 5: Post-Review Test Gate — PrizmKit Test
216
216
 
217
217
  After valid code-review PASS and all configured Browser/Manual verification, run the mandatory test gate against the same artifact root:
218
218
 
@@ -220,18 +220,18 @@ After valid code-review PASS and all configured Browser/Manual verification, run
220
220
  /prizmkit-test scope=this-change artifact_dir=.prizmkit/bugfix/{{BUG_ID}}/
221
221
  ```
222
222
 
223
- Read the sibling `test-report.md` and `test-result.json`; they must agree. `TEST_PASS` proceeds to the fix report and retrospective. For `TEST_NEEDS_FIXES`, classify the concrete correction into an allowed `repair_scope`; safe in-budget repair re-enters Implement through the helper and requires fresh Review, configured verification, and Test, while unknown/unsafe scope or exhausted budget blocks. `TEST_BLOCKED` preserves its blocker and records `WORKFLOW_BLOCKED`. Do not require a manifest/attestation package, reinterpret the testing result as a runtime crash, or recreate the skill's internal loops.
223
+ Read the sibling `test-report.md` and `test-result.json`; they must agree. `TEST_PASS` with `production_changed=false` proceeds to the fix report and retrospective; `TEST_PASS` with consistent `production_changed=true` keeps `prizmkit-test` incomplete and re-enters `prizmkit-code-review` in delta mode within the shared 3-round outer repair budget (`repair_round_limit` records `WORKFLOW_BLOCKED`). For `TEST_NEEDS_FIXES`, classify the concrete correction into an allowed `repair_scope`; safe in-budget repair re-enters Implement through the helper and requires fresh Review, configured verification, and Test, while unknown/unsafe scope or exhausted budget blocks. `TEST_BLOCKED` preserves its blocker and records `WORKFLOW_BLOCKED`. Do not require a manifest/attestation package, reinterpret the testing result as a runtime crash, or recreate the skill's internal loops.
224
224
 
225
- Only `TEST_PASS` may proceed to commit.
225
+ Only `TEST_PASS` with `production_changed=false` may proceed to commit.
226
226
 
227
227
  {{IF_VERIFICATION_MANUAL_OR_HYBRID}}
228
228
  **MANUAL/HYBRID AUTOMATION**: This pipeline session is always headless. Execute manual/hybrid verification with available deterministic browser, CLI/API, fixture, or test tooling. Document the method and verification basis, then continue through `/prizmkit-retrospective` and `/prizmkit-committer`; never ask for UAT, wait for a user, or stop as `partial`.
229
- - Do not write `verifying` to the external bug-list status field; fine-grained verification is session/checkpoint metadata, while the external list status remains one of `pending`, `in_progress`, `completed`, `failed`, `skipped`, or `needs_info`.
229
+ - Do not write `verifying` to the external bug-list status field; fine-grained verification is session/checkpoint metadata, while the external list status remains one of `pending`, `in_progress`, `completed`, `failed`, `skipped`, or `needs_info` (plus runtime-managed `auto_skipped`).
230
230
  {{END_IF_VERIFICATION_MANUAL_OR_HYBRID}}
231
231
 
232
232
  ---
233
233
 
234
- ### Phase 4: Report, Retrospective & Runtime Commit Handoff
234
+ ### Phase 6: Report, Retrospective & Runtime Commit Handoff
235
235
 
236
236
  **a. Fix report before commit preparation**
237
237