prizmkit 1.1.104 → 1.1.106

package/bundled/VERSION.json

@@ -1,5 +1,5 @@
 {
-  "frameworkVersion": "1.1.104",
-  "bundledAt": "2026-07-06T16:42:02.971Z",
-  "bundledFrom": "05c5c13"
+  "frameworkVersion": "1.1.106",
+  "bundledAt": "2026-07-07T02:54:30.507Z",
+  "bundledFrom": "8e4fe49"
 }

package/bundled/dev-pipeline/README.md

@@ -678,10 +678,9 @@ cli.py feature run command
   +- AI CLI session               # claude -p "$(cat prompt)" --dangerously-skip-permissions
   |   |                           # cbc --print -y < prompt
   |   |                           # codex --ask-for-approval never --sandbox danger-full-access exec --cd "$PROJECT_ROOT" --skip-git-repo-check - < prompt
-  |   +- prizm-dev-team           # Multi-agent team implements the feature
-  |       +- Orchestrator         # Main: init, context, plan, retrospective, commit
-  |       +- Dev x N              # Implementation with TDD
-  |       +- Reviewer             # Analyze + code review
+  |   +- session orchestrator     # Main: init, context, plan, direct implementation, review gate, retrospective, commit
+  |       +- Critic               # Optional plan challenge
+  |       +- Reviewer             # Code review via /prizmkit-code-review
   |
   +- check-session-status.py      # Parse session outcome
   +- update-feature-status.py     # Update feature state (completed/failed/retry)
@@ -696,8 +695,8 @@ The bootstrap prompt adapts based on feature complexity:
 | Tier | Template | Agents | Use Case |
 |------|----------|--------|----------|
 | Tier 1 (lite) | `bootstrap-tier1.md` | Single agent handles everything | Low complexity features |
-| Tier 2 (standard) | `bootstrap-tier2.md` | Orchestrator + Dev + Reviewer subagents | Medium complexity |
-| Tier 3 (full) | `bootstrap-tier3.md` | Full team with spec/plan/analyze/implement/review phases | High/critical complexity |
+| Tier 2 (standard) | `bootstrap-tier2.md` | Orchestrator implements directly + optional Critic/Reviewer | Medium/high complexity |
+| Tier 3 (full) | `bootstrap-tier3.md` | Orchestrator implements directly + Critic/Reviewer review phases | High/critical complexity |
 
 **Self-evolve mode:** Uses Tier 3 template with additional framework guardrails for developing the PrizmKit framework itself.
 
@@ -919,7 +918,7 @@ cli.py bugfix run command
 | Claude Code | `.claude/agents/prizm-dev-team-*.md` | `.claude/team-info.json` |
 | CodeBuddy | `.codebuddy/agents/prizm-dev-team-*.md` | `~/.codebuddy/teams/prizm-dev-team/config.json` |
 
-Agent files are 2 types: `prizm-dev-team-dev.md` (implementation) and `prizm-dev-team-reviewer.md` (review). Paths are auto-resolved by `generate-bootstrap-prompt.py`.
+Agent files include Reviewer/Critic definitions used by feature prompt flows. `prizm-dev-team-dev.md` remains available for targeted fix/specialized handoffs such as the internal `/prizmkit-code-review` fix loop, but top-level feature implementation is performed directly by the main orchestrator with `/prizmkit-implement`. Paths are auto-resolved by `generate-bootstrap-prompt.py`.
 
 ---
 

package/bundled/dev-pipeline/assets/prizm-dev-team-integration.md

@@ -2,7 +2,7 @@
 
 ## Overview
 
-dev-pipeline drives the prizm-dev-team multi-agent team through an outer shell loop. Each iteration spawns a new AI CLI session with a bootstrap prompt that instructs the agent to create and orchestrate the team for one feature.
+dev-pipeline drives feature sessions through the unified Python outer loop. Each iteration spawns a new AI CLI session with a bootstrap prompt that makes the main session orchestrator responsible for implementation, while Critic and Reviewer agents remain available for challenge/review phases.
 
 ## Architecture
 
@@ -19,7 +19,7 @@ dev-pipeline (outer loop)
             ├── Phase 1-2: Context snapshot + Specify + Plan (Orchestrator)
             ├── Phase 3: Analyze (Reviewer agent) [tier2] / Phase 4: Analyze [tier3]
             ├── Phase 3.5: Plan Challenge (Critic agent(s), 1 or 3 parallel) [tier2+]
-            ├── Phase 4: Implement (Dev agent) [tier2] / Phase 5: Implement [tier3]
+            ├── Phase 4: Implement (main orchestrator runs /prizmkit-implement directly)
             ├── Phase 5: Review (Reviewer agent)
             └── Phase 6: Retrospective & Commit (Orchestrator)
             │
@@ -30,7 +30,7 @@ dev-pipeline (outer loop)
 
 | Agent | Definition Path | Type |
 |-------|----------------|------|
-| Dev | `.claude/agents/prizm-dev-team-dev.md` (or `.codebuddy/agents/`) | prizm-dev-team-dev |
+| Dev | `.claude/agents/prizm-dev-team-dev.md` (or `.codebuddy/agents/`) | prizm-dev-team-dev; reserved for targeted fix/specialized handoffs, not top-level feature implementation |
 | Reviewer | `.claude/agents/prizm-dev-team-reviewer.md` (or `.codebuddy/agents/`) | prizm-dev-team-reviewer |
 | Critic | `.claude/agents/prizm-dev-team-critic.md` (or `.codebuddy/agents/`) | prizm-dev-team-critic |
 

package/bundled/dev-pipeline/prizmkit_runtime/gitops.py

@@ -565,7 +565,6 @@ def branch_ensure_return_plan(context: BranchContext) -> GitCommandPlan:
             ),
             _git("stash", "push", "-m", "pipeline-ensure-return-stash", allow_failure=True, description="stash checkout blockers"),
             _git("checkout", context.base_branch, allow_failure=True, description="return to original branch"),
-            _git("stash", "pop", allow_failure=True, description="restore stashed checkout blockers"),
         ),
         never_fails=True,
         notes=("Always returns success to support trap/finally cleanup paths.",),

package/bundled/dev-pipeline/prizmkit_runtime/runner_bookkeeping.py

@@ -7,10 +7,11 @@ legacy ordering around branch creation and final status updates.
 
 from __future__ import annotations
 
+from collections.abc import Sequence
 from dataclasses import dataclass
 from pathlib import Path
 
-from .gitops import GitCommandResult, git_status_safe, run_git_command
+from .gitops import HIDDEN_TOOL_WORKTREE_EXCLUDES, GitCommandResult, git_status_safe, run_git_command
 from .runner_models import RunnerFamily
 
 
@@ -27,25 +28,95 @@ class BookkeepingResult:
         return self.result is None or self.result.return_code == 0
 
 
-def has_workspace_changes(project_root: Path) -> bool:
-    """Return true when git status has non-hidden-tool worktree changes."""
-    return bool(git_status_safe(project_root).strip())
+def _bookkeeping_paths(project_root: Path, family: RunnerFamily, item_id: str) -> tuple[str, ...]:
+    return (
+        _repo_path(project_root, family.plan_path),
+        _repo_path(project_root, family.state_dir / item_id / "status.json"),
+    )
+
+
+def _repo_path(project_root: Path, path: Path) -> str:
+    try:
+        return path.resolve().relative_to(project_root.resolve()).as_posix()
+    except ValueError:
+        return path.as_posix()
+
+
+def _changed_paths_in(project_root: Path, paths: Sequence[str]) -> tuple[str, ...]:
+    result = run_git_command(project_root, ("status", "--porcelain", "--", *paths))
+    if result.return_code != 0:
+        return ()
+    changed: list[str] = []
+    for line in result.stdout.splitlines():
+        path = line[3:].strip()
+        if " -> " in path:
+            path = path.split(" -> ", 1)[1].strip()
+        if path:
+            changed.append(path)
+    return tuple(changed)
+
+
+def commit_bookkeeping_changes(
+    project_root: Path,
+    family: RunnerFamily,
+    item_id: str,
+    message: str,
+) -> BookkeepingResult:
+    """Stage and commit only runner-owned bookkeeping paths when present."""
+    paths = _changed_paths_in(project_root, _bookkeeping_paths(project_root, family, item_id))
+    if not paths:
+        return BookkeepingResult(attempted=False, committed=False)
+    run_git_command(project_root, ("add", "-A", "--", *paths))
+    result = run_git_command(project_root, ("commit", "--no-verify", "-m", message))
+    return BookkeepingResult(attempted=True, committed=result.return_code == 0, result=result)
 
 
-def commit_all_changes(project_root: Path, message: str) -> BookkeepingResult:
-    """Stage and commit current workspace changes when present."""
-    if not has_workspace_changes(project_root):
+def commit_task_branch_changes(
+    project_root: Path,
+    family: RunnerFamily,
+    item_id: str,
+    status: str,
+) -> BookkeepingResult:
+    """Commit task branch code, artifacts, and status without switching branches."""
+    message = f"chore({item_id}): save {family.kind} {status} state"
+    if status == "completed":
+        message = f"chore({item_id}): complete {family.kind} task"
+    if not git_status_safe(project_root).strip():
         return BookkeepingResult(attempted=False, committed=False)
-    run_git_command(project_root, ("add", "-A", "--", "."))
+    run_git_command(project_root, ("add", "-A", "--", ".", *HIDDEN_TOOL_WORKTREE_EXCLUDES))
     result = run_git_command(project_root, ("commit", "--no-verify", "-m", message))
     return BookkeepingResult(attempted=True, committed=result.return_code == 0, result=result)
 
 
+def commit_wip_changes(
+    project_root: Path,
+    family: RunnerFamily,
+    item_id: str,
+    status: str,
+) -> BookkeepingResult:
+    """Commit all visible task-branch changes as preserved WIP."""
+    if not git_status_safe(project_root).strip():
+        return BookkeepingResult(attempted=False, committed=False)
+    run_git_command(project_root, ("add", "-A", "--", ".", *HIDDEN_TOOL_WORKTREE_EXCLUDES))
+    result = run_git_command(
+        project_root,
+        (
+            "commit",
+            "--no-verify",
+            "-m",
+            f"wip({item_id}): preserve {family.kind} {status} state",
+            "-m",
+            "Pipeline preserved task branch work without merging it to the base branch.",
+        ),
+    )
+    return BookkeepingResult(attempted=True, committed=result.return_code == 0, result=result)
+
+
 def commit_pre_branch_bookkeeping(project_root: Path, family: RunnerFamily, item_id: str) -> BookkeepingResult:
-    """Commit status/list changes before creating the item dev branch."""
-    return commit_all_changes(project_root, f"chore({item_id}): mark in_progress")
+    """Commit in-progress status on the current task branch."""
+    return commit_bookkeeping_changes(project_root, family, item_id, f"chore({item_id}): mark in_progress")
 
 
 def commit_final_bookkeeping(project_root: Path, family: RunnerFamily, item_id: str, status: str) -> BookkeepingResult:
-    """Commit final list/state changes after returning to the base branch."""
-    return commit_all_changes(project_root, f"chore({item_id}): update {family.kind} status {status}")
+    """Commit final runner-owned status changes on the current task branch."""
+    return commit_bookkeeping_changes(project_root, family, item_id, f"chore({item_id}): update {family.kind} status {status}")

package/bundled/dev-pipeline/prizmkit_runtime/runner_classification.py

@@ -43,10 +43,18 @@ def classify_session(
 ) -> SessionClassification:
     """Classify raw AI session signals with durable success precedence."""
     fingerprint = progress_fingerprint(project_root, prompt)
-    if _has_commit_since(project_root, base_head):
-        return SessionClassification("success", reason="commit_since_base", progress_fingerprint=fingerprint)
+    has_commit_since_base = _has_commit_since(project_root, base_head)
+
+    if _is_terminal_success(result) and _has_completed_workflow_artifact(prompt):
+        return SessionClassification(
+            "success",
+            reason="terminal_success_with_completed_checkpoint",
+            progress_fingerprint=fingerprint,
+        )
 
     if _is_context_overflow(result):
+        if has_commit_since_base:
+            return SessionClassification("success", reason="commit_since_base", progress_fingerprint=fingerprint)
         no_progress_count = _next_no_progress_count(previous_fingerprint, fingerprint, previous_no_progress_count)
         if no_progress_count >= 2:
             return SessionClassification(
@@ -72,6 +80,8 @@ def classify_session(
         return SessionClassification("timed_out", reason="timeout", progress_fingerprint=fingerprint)
     if result.exit_code not in (0, None):
         return SessionClassification("crashed", reason=f"exit_code={result.exit_code}", progress_fingerprint=fingerprint)
+    if has_commit_since_base:
+        return SessionClassification("success", reason="commit_since_base", progress_fingerprint=fingerprint)
     if _has_workspace_changes(project_root):
         _auto_commit_workspace(project_root)
         return SessionClassification("success", reason="auto_commit_uncommitted_changes", progress_fingerprint=fingerprint)
@@ -100,6 +110,32 @@ def progress_fingerprint(project_root: Path, prompt: PromptGenerationResult | No
     }
 
 
+def _is_terminal_success(result: AISessionResult) -> bool:
+    return (
+        result.exit_code in (0, None)
+        and not result.timed_out
+        and result.termination_reason not in {"timeout", "stale_session"}
+        and not result.fatal_error_code
+        and not result.progress_summary.fatal_error_code
+        and result.progress_summary.result_subtype == "success"
+    )
+
+
+def _has_completed_workflow_artifact(prompt: PromptGenerationResult | None) -> bool:
+    if prompt is None:
+        return False
+    if prompt.checkpoint_path and prompt.checkpoint_path.is_file():
+        try:
+            data = json.loads(prompt.checkpoint_path.read_text(encoding="utf-8"))
+        except (OSError, json.JSONDecodeError):
+            data = {}
+        if isinstance(data, dict) and _checkpoint_cursor(data) == "complete":
+            return True
+    if prompt.artifact_path and (prompt.artifact_path / "completion-summary.json").is_file():
+        return True
+    return False
+
+
 def _next_no_progress_count(previous: object | None, current: object, count: int) -> int:
     if previous and previous == current:
         return max(0, count) + 1