prizmkit 1.1.120 → 1.1.121

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

@@ -381,6 +381,33 @@ def _set_continuation_common(status_data, args, session_id):
         status_data["last_session_id"] = session_id
 
 
+def _set_ai_error_continuation_metadata(status_data, args, session_id, reason):
+    """Preserve durable task/session state for any retryable launched AI error."""
+    if args.active_dev_branch:
+        status_data["active_dev_branch"] = args.active_dev_branch
+    if args.base_branch:
+        status_data["base_branch"] = args.base_branch
+    status_data["continuation_pending"] = True
+    status_data["continuation_reason"] = reason
+    status_data["continuation_count"] = status_data.get("continuation_count", 0) + 1
+    status_data["previous_session_id"] = session_id
+    status_data["last_fatal_error_code"] = args.last_fatal_error_code or reason
+    if args.continuation_summary_path:
+        status_data["continuation_summary_path"] = args.continuation_summary_path
+    if args.no_progress_count is not None:
+        status_data["no_progress_count"] = max(0, args.no_progress_count)
+    fingerprint = _load_progress_fingerprint(args.progress_fingerprint)
+    if fingerprint is not None:
+        status_data["last_progress_fingerprint"] = fingerprint
+    status_data["resume_from_phase"] = None
+    if session_id:
+        sessions = status_data.get("sessions", [])
+        if session_id not in sessions:
+            sessions.append(session_id)
+        status_data["sessions"] = sessions
+        status_data["last_session_id"] = session_id
+
+
 def set_context_overflow_continuation_metadata(status_data, args, session_id):
     """Record continuation metadata for a context-overflow outcome in status.json."""
     status_data["context_overflow_count"] = status_data.get("context_overflow_count", 0) + 1
@@ -848,6 +875,18 @@ def action_update(args, feature_list_path, state_dir):
 
     # Track what status we write to feature-list.json
     new_status = current_list_status
+    if args.active_dev_branch:
+        fs["active_dev_branch"] = args.active_dev_branch
+    if args.base_branch:
+        fs["base_branch"] = args.base_branch
+    if args.last_fatal_error_code:
+        fs["last_fatal_error_code"] = args.last_fatal_error_code
+    if args.continuation_summary_path:
+        fs["continuation_summary_path"] = args.continuation_summary_path
+    fingerprint = _load_progress_fingerprint(args.progress_fingerprint)
+    if fingerprint is not None:
+        fs["last_progress_fingerprint"] = fingerprint
+
 
     if session_status == "success":
         fs["infra_error_count"] = 0
@@ -923,10 +962,14 @@ def action_update(args, feature_list_path, state_dir):
 
         if infra_error_count >= max_infra_retries:
             new_status = "failed"
+            fs["continuation_pending"] = False
+            fs["continuation_reason"] = session_status
+            fs["needs_attention"] = True
             if session_id:
                 fs["last_failed_session_id"] = session_id
         else:
-            new_status = "pending"
+            new_status = "in_progress"
+            _set_ai_error_continuation_metadata(fs, args, session_id, session_status)
 
         err = update_feature_in_list(feature_list_path, feature_id, new_status)
         if err:
@@ -955,8 +998,12 @@ def action_update(args, feature_list_path, state_dir):
 
         if fs["retry_count"] >= max_retries:
             new_status = "failed"
+            fs["continuation_pending"] = False
+            fs["continuation_reason"] = session_status
+            fs["needs_attention"] = True
         else:
-            new_status = "pending"
+            new_status = "in_progress"
+            _set_ai_error_continuation_metadata(fs, args, session_id, session_status)
 
         fs["resume_from_phase"] = None
         if session_id:
@@ -969,7 +1016,7 @@ def action_update(args, feature_list_path, state_dir):
             error_out("Failed to update .prizmkit/plans/feature-list.json: {}".format(err))
             return
 
-    if session_status == "success" and session_id:
+    if session_id:
         sessions = fs.get("sessions", [])
         if session_id not in sessions:
             sessions.append(session_id)
@@ -1013,8 +1060,9 @@ def action_update(args, feature_list_path, state_dir):
         summary["degraded_reason"] = session_status
         summary["restart_policy"] = "finalization_retry"
     elif session_status == "infra_error":
-        summary["restart_policy"] = "infra_retry"
+        summary["restart_policy"] = "in_process_ai_error_continuation" if new_status == "in_progress" else "needs_attention"
         summary["infra_error_count"] = fs.get("infra_error_count", 0)
+        summary.update(continuation_metadata_summary(fs))
         summary["artifacts_preserved"] = True
     elif session_status == "context_overflow":
         summary["restart_policy"] = "context_overflow_continuation"
@@ -1025,7 +1073,8 @@ def action_update(args, feature_list_path, state_dir):
         summary.update(continuation_metadata_summary(fs))
         summary["artifacts_preserved"] = True
     elif session_status != "success":
-        summary["restart_policy"] = "preserve_and_retry"
+        summary["restart_policy"] = "in_process_ai_error_continuation" if new_status == "in_progress" else "needs_attention"
+        summary.update(continuation_metadata_summary(fs))
         summary["artifacts_preserved"] = True
 
     print(json.dumps(summary, indent=2, ensure_ascii=False))

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

@@ -275,6 +275,33 @@ def _set_continuation_common(status_data, args, session_id):
         status_data["last_session_id"] = session_id
 
 
+def _set_ai_error_continuation_metadata(status_data, args, session_id, reason):
+    """Preserve durable task/session state for any retryable launched AI error."""
+    if args.active_dev_branch:
+        status_data["active_dev_branch"] = args.active_dev_branch
+    if args.base_branch:
+        status_data["base_branch"] = args.base_branch
+    status_data["continuation_pending"] = True
+    status_data["continuation_reason"] = reason
+    status_data["continuation_count"] = status_data.get("continuation_count", 0) + 1
+    status_data["previous_session_id"] = session_id
+    status_data["last_fatal_error_code"] = args.last_fatal_error_code or reason
+    if args.continuation_summary_path:
+        status_data["continuation_summary_path"] = args.continuation_summary_path
+    if args.no_progress_count is not None:
+        status_data["no_progress_count"] = max(0, args.no_progress_count)
+    fingerprint = _load_progress_fingerprint(args.progress_fingerprint)
+    if fingerprint is not None:
+        status_data["last_progress_fingerprint"] = fingerprint
+    status_data["resume_from_phase"] = None
+    if session_id:
+        sessions = status_data.get("sessions", [])
+        if session_id not in sessions:
+            sessions.append(session_id)
+        status_data["sessions"] = sessions
+        status_data["last_session_id"] = session_id
+
+
 def set_context_overflow_continuation_metadata(status_data, args, session_id):
     """Record continuation metadata for a context-overflow outcome in status.json."""
     status_data["context_overflow_count"] = status_data.get("context_overflow_count", 0) + 1
@@ -460,6 +487,18 @@ def action_update(args, refactor_list_path, state_dir):
 
     # Track what status we write to refactor-list.json
     new_status = current_list_status
+    if args.active_dev_branch:
+        rs["active_dev_branch"] = args.active_dev_branch
+    if args.base_branch:
+        rs["base_branch"] = args.base_branch
+    if args.last_fatal_error_code:
+        rs["last_fatal_error_code"] = args.last_fatal_error_code
+    if args.continuation_summary_path:
+        rs["continuation_summary_path"] = args.continuation_summary_path
+    fingerprint = _load_progress_fingerprint(args.progress_fingerprint)
+    if fingerprint is not None:
+        rs["last_progress_fingerprint"] = fingerprint
+
 
     if session_status == "success":
         rs["infra_error_count"] = 0
@@ -503,10 +542,14 @@ def action_update(args, refactor_list_path, state_dir):
 
         if infra_error_count >= max_infra_retries:
             new_status = "failed"
+            rs["continuation_pending"] = False
+            rs["continuation_reason"] = session_status
+            rs["needs_attention"] = True
             if session_id:
                 rs["last_failed_session_id"] = session_id
         else:
-            new_status = "pending"
+            new_status = "in_progress"
+            _set_ai_error_continuation_metadata(rs, args, session_id, session_status)
 
         err = update_refactor_in_list(refactor_list_path, refactor_id, new_status)
         if err:
@@ -532,27 +575,25 @@ def action_update(args, refactor_list_path, state_dir):
     else:
         rs["retry_count"] = rs.get("retry_count", 0) + 1
 
-        cleaned = cleanup_refactor_artifacts(
-            state_dir=state_dir,
-            refactor_id=refactor_id,
-            project_root=args.project_root,
-        )
-
         if rs["retry_count"] >= max_retries:
             new_status = "failed"
+            rs["continuation_pending"] = False
+            rs["continuation_reason"] = session_status
+            rs["needs_attention"] = True
         else:
-            new_status = "pending"
+            new_status = "in_progress"
+            _set_ai_error_continuation_metadata(rs, args, session_id, session_status)
 
         rs["resume_from_phase"] = None
-        rs["sessions"] = []
-        rs["last_session_id"] = None
+        if session_id:
+            rs["last_failed_session_id"] = session_id
 
         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))
             return
 
-    if session_status == "success" and session_id:
+    if session_id:
         sessions = rs.get("sessions", [])
         if session_id not in sessions:
             sessions.append(session_id)
@@ -594,8 +635,9 @@ def action_update(args, refactor_list_path, state_dir):
         summary["degraded_reason"] = session_status
         summary["restart_policy"] = "finalization_retry"
     elif session_status == "infra_error":
-        summary["restart_policy"] = "infra_retry"
+        summary["restart_policy"] = "in_process_ai_error_continuation" if new_status == "in_progress" else "needs_attention"
         summary["infra_error_count"] = rs.get("infra_error_count", 0)
+        summary.update(continuation_metadata_summary(rs))
         summary["artifacts_preserved"] = True
     elif session_status == "context_overflow":
         summary["restart_policy"] = "context_overflow_continuation"
@@ -606,8 +648,9 @@ def action_update(args, refactor_list_path, state_dir):
         summary.update(continuation_metadata_summary(rs))
         summary["artifacts_preserved"] = True
     elif session_status != "success":
-        summary["restart_policy"] = "full_restart"
-        summary["cleanup_performed"] = cleaned
+        summary["restart_policy"] = "in_process_ai_error_continuation" if new_status == "in_progress" else "needs_attention"
+        summary.update(continuation_metadata_summary(rs))
+        summary["artifacts_preserved"] = True
 
     print(json.dumps(summary, indent=2, ensure_ascii=False))
 

package/bundled/dev-pipeline/templates/bootstrap-tier3.md

@@ -240,7 +240,7 @@ For each Verification Gate from the Task Contract, write one evidence line to Im
 
 Review the completed implementation before the scoped feature test gate runs. Use `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/` so the code-review skill can read spec, plan, context snapshot, implementation log, and current diff artifacts. Do not require or read a prior `/prizmkit-test` report here; test generation and test-fix looping happen in the next phase.
 
-Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The code-review skill must launch Reviewer in `current-workspace / active-checkout / no-worktree` mode so it reviews the active checkout rather than an isolated copy; do not request worktree isolation or use `.claude/worktrees/...` / `.prizmkit/state/worktrees/...`. Do not set `USE_WORKTREE` for reviewer launches.
+Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The skill chooses a platform-supported independent Reviewer strategy and verifies that the Reviewer's snapshot is equivalent to the current workspace under its skill-owned workspace protocol. Do not prescribe Agent launch parameters from this bootstrap prompt.
 
 The skill runs an internal review-fix loop (Reviewer Agent → filter → orchestrator fix, max 3 rounds) and writes `review-report.md` to the artifact directory.
 
@@ -253,7 +253,7 @@ If GATE:MISSING:
 - Do not re-run `/prizmkit-code-review` in an unbounded report-repair loop.
 - Perform one bounded status check; retry at most once: inspect the skill output, `review-report.md` path, and any Reviewer agent spawn messages.
 - If the missing report is caused by Reviewer agent spawn or review skill error, retry `/prizmkit-code-review` at most once only if it appears transient.
-- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then either perform a safe inline fallback review (spec/plan/diff/tests → write `review-report.md` with `## Verdict`) or stop with a clear recovery failure.
+- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then stop with a clear recovery failure. Do not synthesize a passing report through Main-Agent self-review.
 
 Read `review-report.md` and check the Verdict:
 - `PASS` → proceed to next phase

package/bundled/dev-pipeline/templates/bugfix-bootstrap-prompt.md

@@ -180,6 +180,7 @@ After implement completes, verify:
 If `FAST_PATH=true` (≤ 2 tasks, obvious root cause), skip this phase entirely and mark checkpoint step `prizmkit-code-review` as `skipped` before continuing.
 
 Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/bugfix/{{BUG_ID}}/`:
+- The skill chooses a platform-supported independent Reviewer strategy and verifies a workspace-equivalent snapshot under its skill-owned protocol
 - The skill runs an internal review-fix loop (Reviewer → filter → orchestrator fix, max 3 rounds) and writes `review-report.md`
 - `review-report.md` must contain `## Verdict`
 
@@ -191,7 +192,7 @@ After `/prizmkit-code-review` returns, verify the review report:
 If GATE:MISSING:
 - Do not re-run `/prizmkit-code-review` in an unbounded report-repair loop.
 - Perform one bounded status check; retry `/prizmkit-code-review` at most once only if the missing report appears caused by a transient team/config/lock error.
-- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then either perform a safe inline fallback review (spec/plan/diff/tests → write `review-report.md` with `## Verdict`) or stop with a clear recovery failure.
+- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then stop with a clear recovery failure. Do not replace the independent Reviewer with Main-Agent self-review.
 
 Read `review-report.md` and check the Verdict:
 - `PASS` → proceed

package/bundled/dev-pipeline/templates/refactor-bootstrap-prompt.md

@@ -139,7 +139,7 @@ Apply the browser behavior preservation protocol where UI behavior can be affect
 
 ### Phase 4: Review — Code Review & Behavior Verification
 
-Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/refactor/{{REFACTOR_ID}}/` directly from the main orchestrator.
+Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/refactor/{{REFACTOR_ID}}/` directly from the main orchestrator. The skill loads its skill-owned `references/reviewer-agent-prompt.md` and `references/reviewer-workspace-protocol.md`, chooses a platform-supported independent Reviewer strategy, and verifies a workspace-equivalent snapshot.
 
 Review requirements:
 - verify refactor quality against spec.md and plan.md;
@@ -147,7 +147,7 @@ Review requirements:
 - run or confirm relevant tests;
 - write `.prizmkit/refactor/{{REFACTOR_ID}}/review-report.md` with `## Verdict`.
 
-If the report is missing after one bounded status check, write `failure-log.md` and either perform a safe inline fallback review or stop with a clear recovery failure.
+If the report is missing after one bounded status check, write `failure-log.md` and stop with a clear recovery failure. Do not replace the independent Reviewer with Main-Agent self-review.
 
 **Checkpoint update**: set step `prizmkit-code-review` to `completed` in `{{CHECKPOINT_PATH}}`.
 

package/bundled/dev-pipeline/templates/sections/bugfix-phase-review.md

@@ -12,9 +12,9 @@ If `FAST_PATH=true`, skip this phase intentionally:
 
 Then continue to the next checkpoint step.
 
-Otherwise, run `/prizmkit-code-review` with `artifact_dir=.prizmkit/bugfix/{{BUG_ID}}/`. The code-review skill loads its skill-owned `references/reviewer-agent-prompt.md` prompt and launches the internal reviewer in `current-workspace / active-checkout / no-worktree` mode.
+Otherwise, run `/prizmkit-code-review` with `artifact_dir=.prizmkit/bugfix/{{BUG_ID}}/`. The skill loads its skill-owned Reviewer prompt and workspace protocol, chooses a platform-supported subagent strategy, and verifies that the Reviewer's snapshot is equivalent to the current workspace before review.
 
-The skill runs its internal Reviewer Agent → filter → orchestrator fix loop and writes `review-report.md`. Do not request worktree isolation or use `.claude/worktrees/...` / `.prizmkit/state/worktrees/...`; outer runner `USE_WORKTREE` settings do not permit review subagent isolation.
+The skill runs its internal Reviewer Agent → filter → orchestrator fix loop and writes `review-report.md`. Do not prescribe Agent launch parameters from this pipeline template; the review skill owns strategy selection and workspace-equivalence verification.
 
 **Gate Check — Review Report**:
 
@@ -24,6 +24,6 @@ The skill runs its internal Reviewer Agent → filter → orchestrator fix loop
 
 - `PASS` → proceed.
 - `NEEDS_FIXES` → the skill exhausted its internal rounds; log remaining findings and proceed only when they do not block acceptance criteria.
-- Missing report → one bounded retry or safe inline fallback review, then write `failure-log.md` if still blocked.
+- Missing report → one bounded retry, then write `failure-log.md` and stop if the independent review protocol remains blocked; do not replace it with Main-Agent self-review.
 
 **Checkpoint update**: Set step `prizmkit-code-review` to `completed` when review is complete, or `skipped` for FAST_PATH.

package/bundled/dev-pipeline/templates/sections/phase-review-agent.md

@@ -2,9 +2,9 @@
 
 Review the completed implementation before the scoped feature test gate runs. Use `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/` so the code-review skill can read spec, plan, context snapshot, implementation log, and current diff artifacts. Do not require or read a prior `/prizmkit-test` report here; test generation and test-fix looping happen in the next phase.
 
-Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The code-review skill loads its skill-owned `references/reviewer-agent-prompt.md` prompt and launches the internal reviewer in `current-workspace / active-checkout / no-worktree` mode.
+Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The skill loads its skill-owned Reviewer prompt and workspace protocol, chooses a platform-supported subagent strategy, and verifies that the Reviewer's snapshot is equivalent to the current workspace before review.
 
-The skill runs an internal review-fix loop (Reviewer Agent → filter → orchestrator fix, max 3 rounds) and writes `review-report.md` to the artifact directory. Do not request worktree isolation or use `.claude/worktrees/...` / `.prizmkit/state/worktrees/...`; outer runner `USE_WORKTREE` settings do not permit review subagent isolation.
+The skill runs an internal review-fix loop (Reviewer Agent → filter → orchestrator fix, max 3 rounds) and writes `review-report.md` to the artifact directory. Do not prescribe Agent launch parameters from this pipeline template; the review skill owns strategy selection and workspace-equivalence verification.
 
 **Gate Check — Review Report**:
 After `/prizmkit-code-review` returns, verify the review report:
@@ -15,7 +15,7 @@ If GATE:MISSING:
 - Do not re-run `/prizmkit-code-review` in an unbounded report-repair loop.
 - Perform one bounded status check; retry at most once: inspect the skill output, `review-report.md` path, and any Reviewer agent spawn messages.
 - If the missing report is caused by team/config/lock errors from the Reviewer agent spawn or review skill error, retry `/prizmkit-code-review` at most once only if it appears transient.
-- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then either perform a safe inline fallback review (spec/plan/diff/tests → write `review-report.md` with `## Verdict`) or stop with a clear recovery failure.
+- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then stop with a clear recovery failure. Do not synthesize a passing report through Main-Agent self-review.
 
 Read `review-report.md` and check the Verdict:
 - `PASS` → proceed to next phase

package/bundled/dev-pipeline/templates/sections/phase-review-full.md

@@ -2,11 +2,11 @@
 
 Review the completed implementation before the scoped feature test gate runs. Use `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/` so the code-review skill can read spec, plan, context snapshot, implementation log, and current diff artifacts. Do not require or read a prior `/prizmkit-test` report here; test generation and test-fix looping happen in the next phase.
 
-Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The code-review skill loads its skill-owned `references/reviewer-agent-prompt.md` prompt and launches the internal reviewer in `current-workspace / active-checkout / no-worktree` mode.
+Run `/prizmkit-code-review` with `artifact_dir=.prizmkit/specs/{{FEATURE_SLUG}}/`. The skill loads its skill-owned Reviewer prompt and workspace protocol, chooses a platform-supported subagent strategy, and verifies that the Reviewer's snapshot is equivalent to the current workspace before review.
 
-If the reviewer spawn fails with team/config/lock errors, retry at most once. If the second attempt fails, do not keep spawning variants and do not enter report-polling loops for review markers; write `failure-log.md` with the spawn error and use the documented inline fallback only when safe.
+If the independent Reviewer cannot start or verify the snapshot because of a transient platform error, retry at most once. If the second attempt fails, do not keep spawning variants or enter report-polling loops; write `failure-log.md` with the observable error and let the skill report the unresolved verification blocker. Do not replace the independent Reviewer with Main-Agent self-review.
 
-The skill runs an internal review-fix loop (Reviewer Agent → filter → orchestrator fix, max 3 rounds) and writes `review-report.md` to the artifact directory. Do not request worktree isolation or use `.claude/worktrees/...` / `.prizmkit/state/worktrees/...`; outer runner `USE_WORKTREE` settings do not permit review subagent isolation.
+The skill runs an internal review-fix loop (Reviewer Agent → filter → orchestrator fix, max 3 rounds) and writes `review-report.md` to the artifact directory. Do not prescribe Agent launch parameters from this pipeline template; the review skill owns strategy selection and workspace-equivalence verification.
 
 **Gate Check — Review Report**:
 After `/prizmkit-code-review` returns, verify the review report:
@@ -17,7 +17,7 @@ If GATE:MISSING:
 - Do not re-run `/prizmkit-code-review` in an unbounded report-repair loop.
 - Perform one bounded status check; retry at most once: inspect the skill output, `review-report.md` path, and any Reviewer agent spawn messages.
 - If the missing report is caused by team/config/lock errors from the Reviewer agent spawn or review skill error, retry `/prizmkit-code-review` at most once only if it appears transient.
-- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then either perform a safe inline fallback review (spec/plan/diff/tests → write `review-report.md` with `## Verdict`) or stop with a clear recovery failure.
+- If the report is still missing after that single check/retry, write `failure-log.md` with the spawn/skill error and last observable state, then stop with a clear recovery failure. Do not synthesize a passing report through Main-Agent self-review.
 
 Read `review-report.md` and check the Verdict:
 - `PASS` → proceed to next phase

package/bundled/dev-pipeline/templates/sections/refactor-phase-review.md

@@ -1,8 +1,8 @@
 ### Phase 3: Review — Code Review & Behavior Verification
 
-Run `/prizmkit-code-review` directly with `artifact_dir=.prizmkit/refactor/{{REFACTOR_ID}}/`. The code-review skill loads its skill-owned `references/reviewer-agent-prompt.md` prompt and launches the internal reviewer in `current-workspace / active-checkout / no-worktree` mode.
+Run `/prizmkit-code-review` directly with `artifact_dir=.prizmkit/refactor/{{REFACTOR_ID}}/`. The skill loads its skill-owned `references/reviewer-agent-prompt.md` and `references/reviewer-workspace-protocol.md`, chooses a platform-supported subagent strategy, and verifies that the Reviewer's snapshot is equivalent to the current workspace before review.
 
-Do not spawn a top-level Reviewer subagent. The code-review skill owns its internal Reviewer Agent → filter → orchestrator fix loop. Do not request worktree isolation or use `.claude/worktrees/...` / `.prizmkit/state/worktrees/...`; outer runner `USE_WORKTREE` settings do not permit review subagent isolation.
+Do not spawn a top-level Reviewer subagent. The code-review skill owns its internal Reviewer Agent → filter → orchestrator fix loop. Do not prescribe Agent launch parameters from this pipeline template; the review skill owns strategy selection and workspace-equivalence verification.
 
 **Gate Check — Review Report**:
 
@@ -12,7 +12,7 @@ Do not spawn a top-level Reviewer subagent. The code-review skill owns its inter
 
 - `PASS` → proceed.
 - `NEEDS_FIXES` → the skill exhausted its internal rounds; log remaining findings and proceed only when they do not block behavior preservation.
-- Missing report → one bounded retry or safe inline fallback review, then write `failure-log.md` if still blocked.
+- Missing report → one bounded retry, then write `failure-log.md` and stop if the independent review protocol remains blocked; do not replace it with Main-Agent self-review.
 
 Verify the review explicitly checks behavior preservation, public API compatibility, imports, tests, and rollback assumptions.
 

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

@@ -1195,7 +1195,7 @@ def _run_update(fl_path, state_dir, feature_id, session_status, session_id="sess
 
 
 class TestInfraErrorUpdate:
-    def test_infra_error_below_budget_keeps_pending_without_consuming_retry(self, tmp_path):
+    def test_infra_error_below_budget_keeps_in_progress_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"])
@@ -1214,12 +1214,12 @@ class TestInfraErrorUpdate:
             max_infra_retries=3,
         )
 
-        assert result["new_status"] == "pending"
+        assert result["new_status"] == "in_progress"
         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"
+        assert result["restart_policy"] == "in_process_ai_error_continuation"
+        assert _read_statuses(fl_path)["F-001"] == "in_progress"
 
         fs = load_feature_status(state_dir, "F-001")
         assert fs["retry_count"] == 2
@@ -1250,7 +1250,7 @@ class TestInfraErrorUpdate:
         assert result["new_status"] == "failed"
         assert result["retry_count"] == 1
         assert result["infra_error_count"] == 3
-        assert result["restart_policy"] == "infra_retry"
+        assert result["restart_policy"] == "needs_attention"
         assert _read_statuses(fl_path)["F-001"] == "failed"
 
         fs = load_feature_status(state_dir, "F-001")

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

@@ -921,7 +921,14 @@ class TestContinuationHandoff:
         args = Namespace(
             output=str(tmp_path / ".prizmkit/state/features/F-123/sessions/F-123-new/bootstrap-prompt.md"),
             state_dir=str(tmp_path / ".prizmkit/state/features"),
-            continuation_mode="context_overflow",
+            continuation_mode="ai_error",
+            failure_classification="ai_runtime_or_provider_error",
+            code_retry_count="1",
+            infra_error_count="2",
+            max_code_retries="3",
+            max_infra_retries="3",
+            no_progress_count="0",
+            progress_fingerprint=json.dumps({"git_diff_fingerprint": "diff", "head_commit": "abc", "checkpoint_cursor": "S03:pending"}),
             previous_session_id="F-123-old",
             continuation_count="2",
             context_overflow_count="2",
@@ -942,11 +949,11 @@ class TestContinuationHandoff:
 
         assert rendered.startswith("NORMAL BOOTSTRAP")
         assert "## Continuation Addendum" in rendered
-        assert "This is an automatic continuation after context overflow." in rendered
+        assert "This is an automatic continuation after the previous launched AI session ended with `ai_runtime_or_provider_error`." in rendered
         assert "This is a fresh headless AI CLI session for the same task." in rendered
         assert "Continue on the current branch/worktree." in rendered
         assert "Do not reset, clean, switch branches, or discard work." in rendered
-        assert "Start with `git status` and `git diff`." in rendered
+        assert "Inspect `git status` and `git diff` before editing." in rendered
         assert "Treat completed/skipped checkpoint steps as already done." in rendered
         assert "Do not read the previous full `session.log` unless explicitly necessary." in rendered
         assert "prefer synthesized artifacts and summaries over logs" in rendered

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

@@ -265,7 +265,14 @@ class TestBugfixContinuationHandoff:
         args = Namespace(
             output=str(tmp_path / ".prizmkit/state/bugfix/B-007/sessions/B-007-new/bootstrap-prompt.md"),
             state_dir=str(tmp_path / ".prizmkit/state/bugfix"),
-            continuation_mode="context_overflow",
+            continuation_mode="ai_error",
+            failure_classification="ai_runtime_or_provider_error",
+            code_retry_count="1",
+            infra_error_count="2",
+            max_code_retries="3",
+            max_infra_retries="3",
+            no_progress_count="0",
+            progress_fingerprint=json.dumps({"git_diff_fingerprint": "diff", "head_commit": "abc", "checkpoint_cursor": "S03:pending"}),
             previous_session_id="B-007-old",
             continuation_count="1",
             context_overflow_count="1",
@@ -279,7 +286,7 @@ class TestBugfixContinuationHandoff:
         )
 
         assert prompt.startswith("BUGFIX PROMPT")
-        assert "This is an automatic continuation after context overflow." in prompt
+        assert "This is an automatic continuation after the previous launched AI session ended with `ai_runtime_or_provider_error`." in prompt
         assert "Do not attempt `/compact` as the recovery mechanism." not in prompt
         assert ("Log " + "monitor") not in prompt
         assert ("log-" + "monitor") not in prompt

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

@@ -1,5 +1,6 @@
 """Tests for generate-refactor-prompt.py continuation behavior."""
 
+import json
 from argparse import Namespace
 from pathlib import Path
 import os
@@ -20,7 +21,14 @@ class TestRefactorContinuationHandoff:
         args = Namespace(
             output=str(tmp_path / ".prizmkit/state/refactor/R-007/sessions/R-007-new/bootstrap-prompt.md"),
             state_dir=str(tmp_path / ".prizmkit/state/refactor"),
-            continuation_mode="context_overflow",
+            continuation_mode="ai_error",
+            failure_classification="ai_runtime_or_provider_error",
+            code_retry_count="1",
+            infra_error_count="2",
+            max_code_retries="3",
+            max_infra_retries="3",
+            no_progress_count="0",
+            progress_fingerprint=json.dumps({"git_diff_fingerprint": "diff", "head_commit": "abc", "checkpoint_cursor": "S03:pending"}),
             previous_session_id="R-007-old",
             continuation_count="1",
             context_overflow_count="1",