prizmkit 1.0.34 → 1.0.45

package/bundled/VERSION.json

@@ -1,5 +1,5 @@
 {
-  "frameworkVersion": "1.0.34",
-  "bundledAt": "2026-03-17T12:56:01.113Z",
-  "bundledFrom": "68ce087"
+  "frameworkVersion": "1.0.45",
+  "bundledAt": "2026-03-18T12:24:34.664Z",
+  "bundledFrom": "6127c1a"
 }

package/bundled/agents/prizm-dev-team-dev.md

@@ -18,7 +18,7 @@ skills: prizmkit-implement, prizmkit-prizm-docs
 
 ### 项目上下文
 
-项目文档在 `.prizm-docs/`。实现前先读 `root.prizm` 了解规则和已知陷阱（TRAPS），修改某模块时读该模块的文档。
+项目文档在 `.prizm-docs/`。实现前先读 `context-snapshot.md`（若存在于 `.prizmkit/specs/###-feature-name/`），其 Section 3 含 Prizm Context、Section 4 含源文件，无需再读 `.prizm-docs/` 或原始源文件。若 snapshot 不存在，则读 `root.prizm` 了解规则和已知陷阱（TRAPS），修改某模块时读该模块的文档。
 
 ### 制品路径
 
@@ -32,17 +32,17 @@ skills: prizmkit-implement, prizmkit-prizm-docs
 1. 按照分配的任务和 plan.md 中的接口设计实现功能模块
 2. 遵循 TDD 方式：先写测试，再实现，再验证
 3. 产出的代码必须通过本模块的单元测试
-4. 发现接口设计歧义时，立即通过 Coordinator 上报 PM（不自行假设）
+4. 发现接口设计歧义时，立即上报给 Orchestrator（不自行假设）
 5. 遵循 `/prizmkit-implement` 工作流：读取 plan.md（含 Tasks section）+ spec.md，逐任务实现
 6. 每个任务完成后**立即**标记 plan.md Tasks section 中的 `[x]`（不批量标记）
-7. 实现前读取 `.prizm-docs/` TRAPS 段避免已知陷阱
+7. 实现前读取 TRAPS 段避免已知陷阱：优先从 `context-snapshot.md` Section 3 获取，若 snapshot 不存在则读 `.prizm-docs/`
 8. 检查点任务须验证构建通过和测试通过后才能继续下一阶段
 9. 顺序任务按序执行，失败则停止；并行 `[P]` 任务可继续
 10. 新建子模块时，生成对应的 `.prizm-docs/` L2 文档
 
 ### 绝不做 (NEVER)
 
-- 不修改 plan.md 中的接口设计（修改需通过 PM）
+- 不修改 plan.md 中的接口设计（修改需通过 Orchestrator）
 - 不修改其他 Dev Agent 负责的模块代码
 - 不进行集成测试（Reviewer 的职责）
 - **不执行任何 git 操作**（git commit / git add / git reset / git push 均禁止 — 由 Orchestrator 通过 /prizmkit-committer 统一提交）
@@ -62,7 +62,7 @@ DEV-06: 不得引入未在任务描述中声明的外部依赖
 DEV-07: 遵循 /prizmkit-implement 工作流
 DEV-08: 每个任务完成后立即标记 plan.md Tasks section [x]
 DEV-09: TDD：先写测试 → 再实现 → 再验证
-DEV-10: 实现每个模块前必须读取 .prizm-docs/ TRAPS 段
+DEV-10: 实现每个模块前必须读取 TRAPS 段：优先从 context-snapshot.md Section 3 获取，无 snapshot 时读 .prizm-docs/
 DEV-11: 检查点任务必须验证构建通过和测试通过
 DEV-12: 新建子模块时生成 L2 .prizm-docs/ 文档
 DEV-13: 禁止执行任何 git 命令（git add/commit/reset/push 全部禁止）
@@ -72,13 +72,13 @@ DEV-14: 若 `npm test` 中存在 pre-existing 失败，不得忽略——必须
 ### 工作流程
 
 1. 接收任务分配
-2. 读取 `.prizm-docs/root.prizm` 和相关模块文档
+2. 读取 `.prizmkit/specs/###-feature-name/context-snapshot.md`（若存在）——Section 3 含 Prizm Context，Section 4 含源文件。若 snapshot 不存在，则读取 `.prizm-docs/root.prizm` 和相关模块文档
 3. 读取 `.prizmkit/specs/###-feature-name/` 中的 `plan.md`（含 Tasks section）、`spec.md`
 4. 对每个分配的任务，按 plan.md Tasks 顺序执行：
-   a. 读取目标文件模块的文档（检查 TRAPS 和 DECISIONS）
+   a. 从 context-snapshot.md 获取目标文件上下文和 TRAPS（若无 snapshot 则读取目标文件模块的文档）
    b. TDD：基于验收标准编写测试 → 实现功能代码 → 运行测试验证
    c. 在 plan.md Tasks section 中标记该任务为 `[x]`
-   d. 发送 STATUS_UPDATE 给 Coordinator
+   d. 发送 STATUS_UPDATE 给 Orchestrator
 5. 如遇检查点任务，验证构建通过和测试通过后才继续
 6. 遇到接口设计歧义，发送 ESCALATION（不自行假设）
 7. 如新建了子模块，生成对应 `.prizm-docs/` L2 文档
@@ -88,25 +88,16 @@ DEV-14: 若 `npm test` 中存在 pre-existing 失败，不得忽略——必须
 
 | 场景 | 策略 |
 |------|------|
-| 接口设计歧义 | 标记 BLOCKED → ESCALATION → 等待 PM 裁定 |
+| 接口设计歧义 | 标记 BLOCKED → ESCALATION → 等待 Orchestrator 裁定 |
 | 单元测试失败 | 最多重试修复 3 次 → 仍失败则 ISSUE_REPORT |
 | 外部依赖不可用 | 使用 Mock → 标注说明 |
-| 任务超出预估 | ESCALATION → 建议 PM 拆分任务 |
+| 任务超出预估 | ESCALATION → 建议 Orchestrator 拆分任务 |
 
 ### 通信规则
 
-允许 Agent 之间直接通信，但关键消息和结论必须通知 Coordinator。
+允许 Agent 之间直接通信，但关键消息和结论必须通知 Orchestrator。
 - 发送 STATUS_UPDATE 汇报每个子任务完成
 - 发送 COMPLETION_SIGNAL 标志所有任务完成
 - 发送 ESCALATION 上报接口歧义或任务阻塞
 - 接收 TASK_ASSIGNMENT 获取分配的工作
 
-### Framework Self-Development Rules (self-evolve mode)
-
-When working in self-evolve mode (developing the PrizmKit framework itself), these additional rules apply:
-
-1. **Skill metadata sync**: If you modify any file in `core/skills/<skill-name>/`, you MUST also update `core/skills/<skill-name>/_metadata.json` to reflect changes (description, version, dependencies).
-2. **Template variable check**: If you modify any file in `dev-pipeline/templates/`, verify all `{{PLACEHOLDER}}` markers are resolvable by checking `dev-pipeline/scripts/generate-bootstrap-prompt.py` for matching entries.
-3. **Pre-completion validation**: Before marking implementation complete, run `node tests/validate-all.js` and fix any failures.
-4. **Bundle protection**: NEVER directly modify files in `create-prizmkit/bundled/`. These are auto-generated by `scripts/bundle.js` — your changes would be overwritten.
-5. **Reload tracking**: If you modify files in `dev-pipeline/scripts/`, `dev-pipeline/templates/`, or `core/skills/` that the pipeline actively uses, note this in your Implementation Log so the orchestrator can set `reload_needed: true`.

package/bundled/agents/prizm-dev-team-reviewer.md

@@ -16,7 +16,7 @@ skills: prizmkit-code-review, prizmkit-analyze, prizmkit-prizm-docs
 
 ### 项目上下文
 
-项目文档在 `.prizm-docs/`。审查前先读 `root.prizm` 了解项目规则（RULES）、模式（PATTERNS）和已知陷阱（TRAPS），需要时读取模块级文档。
+项目文档在 `.prizm-docs/`。审查前先读 `context-snapshot.md`（若存在于 `.prizmkit/specs/###-feature-name/`），其 Section 3 含 Prizm Context（RULES、PATTERNS、TRAPS），无需再读 `.prizm-docs/` 或原始源文件。若 snapshot 不存在，则读 `root.prizm` 了解项目规则（RULES）、模式（PATTERNS）和已知陷阱（TRAPS），需要时读取模块级文档。
 
 ### 制品路径
 
@@ -40,8 +40,8 @@ skills: prizmkit-code-review, prizmkit-analyze, prizmkit-prizm-docs
 ### 绝不做 (NEVER)
 
 - 不编写实现代码（Dev 的职责）
-- 不分解任务（PM 的职责）
-- 不进行任务调度（Coordinator 的职责）
+- 不分解任务（Orchestrator 的职责）
+- 不进行任务调度（Orchestrator 的职责）
 - **不执行任何 git 操作**（git commit / git add / git reset / git push 均禁止）
 - 不使用 TaskCreate/TaskUpdate 创建或修改 Orchestrator 层的任务（Task 工具仅用于内部进度追踪，且任务 ID 在各 agent 子会话中互不共享）
 
@@ -62,21 +62,21 @@ REV-10: 禁止使用 timeout 命令（macOS 不兼容）。运行测试时直接
 
 ### Phase 4 工作流程：交叉校验
 
-**前置条件**: PM 已完成 spec.md / plan.md（含 Tasks section）
+**前置条件**: Orchestrator 已完成 spec.md / plan.md（含 Tasks section）
 
 1. 调用 `/prizmkit-analyze` skill（**不是 CLI 命令**，使用 Skill 工具或 `/prizmkit-analyze` 指令调用）
    - 输入: spec.md, plan.md（含 Tasks section）
    - 6 个检测通道: 重复检测、歧义检测、不完整检测、Prizm 规则对齐、覆盖缺口、不一致性
    - 输出: 一致性分析报告（仅对话输出）
    - 若 Skill 工具不可用，则根据 6 个检测通道手动执行交叉一致性分析
-2. 如发现 CRITICAL 问题，报告给 Coordinator 退回 PM 修复
+2. 如发现 CRITICAL 问题，报告给 Orchestrator 退回修复
 3. 发送 COMPLETION_SIGNAL（含分析结果）
 
 ### Phase 6 工作流程：评审
 
 **前置条件**: Dev 已完成实现，所有任务标记 `[x]`
 
-1. 读取 `.prizm-docs/root.prizm`，重点关注 RULES 和 PATTERNS
+1. 读取 `context-snapshot.md`（若存在），其 Section 3 含 RULES 和 PATTERNS。若 snapshot 不存在，则读 `.prizm-docs/root.prizm`
 2. 运行 `/prizmkit-code-review`（只读）
    - 6 个审查维度: 规格符合度、计划遵循度、代码质量、安全性、一致性、测试覆盖
    - 判定: PASS | PASS WITH WARNINGS | NEEDS FIXES
@@ -109,25 +109,16 @@ REV-10: 禁止使用 timeout 命令（macOS 不兼容）。运行测试时直接
 
 | 场景 | 策略 |
 |------|------|
-| analyze 发现 CRITICAL | 报告 Coordinator → 退回 PM 修复 |
-| code-review 发现 CRITICAL | 报告 Coordinator → 退回 Dev 修复 |
-| 集成测试失败 | 分类严重级别 → ISSUE_REPORT → Coordinator 派发给 Dev |
+| analyze 发现 CRITICAL | 报告 Orchestrator → 退回修复 |
+| code-review 发现 CRITICAL | 报告 Orchestrator → 退回 Dev 修复 |
+| 集成测试失败 | 分类严重级别 → ISSUE_REPORT → Orchestrator 派发给 Dev |
 | 审查发现超过 30 个 | 只保留最严重的 30 个 |
 | Prizm RULES 违规 | 自动标记为 CRITICAL |
 
 ### 通信规则
 
-允许 Agent 之间直接通信，但关键消息和结论必须通知 Coordinator。
+允许 Agent 之间直接通信，但关键消息和结论必须通知 Orchestrator。
 - 发送 COMPLETION_SIGNAL（含判定结果）标志完成
 - 发送 ISSUE_REPORT 报告 CRITICAL 发现
 - 接收 TASK_ASSIGNMENT 获取分配的工作
 
-### Framework Self-Development Review (self-evolve mode)
-
-When reviewing in self-evolve mode (framework is modifying itself), add these review dimensions:
-
-1. **`_metadata.json` ↔ skill directory 1:1 mapping**: Verify every `core/skills/*/` has a `_metadata.json`, and every `_metadata.json` references a valid skill directory. Run `node tests/validate-all.js` to automate this check.
-2. **Template variable completeness**: For modified `dev-pipeline/templates/*.md` files, verify all `{{PLACEHOLDER}}` markers have matching open/close tags and are resolvable by `generate-bootstrap-prompt.py`.
-3. **Agent frontmatter validation**: For modified `core/agents/*.md` files, validate YAML frontmatter contains required fields: `name`, `description`, `tools`. Optionally check `model`, `skills`.
-4. **CI gate execution**: Run `npm run ci` and report the full result. Any failure here is **CRITICAL** severity — the framework must always ship green.
-5. **Bundle safety check**: Verify no files in `create-prizmkit/bundled/` were directly modified (use `git diff --name-only` to check). Direct modifications to bundled assets are always CRITICAL.

package/bundled/dev-pipeline/README.md

@@ -221,12 +221,9 @@ run.sh main loop
   ├─ AI CLI session            # cbc --print -y < prompt (CodeBuddy)
   │   │                        # claude --print -p "$(cat prompt)" --yes (Claude Code)
   │   └─ prizm-dev-team       # Multi-agent team implements the feature
-  │       ├─ Coordinator      # Orchestrates the 10-phase pipeline
-  │       ├─ PM               # Phase 1-4: specify → plan → tasks → analyze
-  │       ├─ Dev x N           # Phase 6: implement with TDD
-  │       ├─ QA               # Phase 7: integration tests + code review
-  │       ├─ Review           # Phase 7: code consistency audit
-  │       └─ Coordinator      # Phase 9: summarize → commit → retrospective
+  │       ├─ Orchestrator    # Main agent: init, plan, schedule, summarize, commit
+  │       ├─ Dev x N          # Implementation with TDD
+  │       └─ Reviewer        # Analyze + code review
   │
   ├─ check-session-status.py  # Parse session outcome
   ├─ update-feature-status.py # Update feature state (completed/failed/retry)
@@ -238,19 +235,19 @@ run.sh main loop
 
 Each AI CLI session drives the prizm-dev-team through these phases. **All phases are mandatory** — the bootstrap prompt enforces sequential execution.
 
-> **Note**: The Coordinator Agent definition describes a simplified **8-phase** view (Phase 0-7) where Phases 1-3 and Phase 5 are consolidated. The bootstrap prompt adapts these phases based on complexity mode (lite/standard/full). The 10-phase breakdown below is the most granular view for pipeline monitoring.
+> **Note**: The bootstrap prompt adapts these phases based on complexity mode (lite/standard/full). The 10-phase breakdown below is the most granular view for pipeline monitoring.
 
 | Phase | Name | Agent | PrizmKit Skills | Artifacts |
 |-------|------|-------|----------------|-----------|
-| 0 | Init | Coordinator | `prizmkit-init` | `.prizm-docs/root.prizm`, `.prizmkit/config.json` |
-| 1 | Specify | PM | `prizmkit-specify`, `prizmkit-clarify` | `.prizmkit/specs/spec.md` |
-| 2 | Plan + Tasks | PM | `prizmkit-plan` | `.prizmkit/specs/plan.md` (含 Tasks section) |
-| 3 | Analyze | PM | `prizmkit-analyze` | Analysis report (no CRITICAL issues) |
-| 4 | Schedule | Coordinator | — | TaskList entries assigned |
+| 0 | Init | Orchestrator | `prizmkit-init` | `.prizm-docs/root.prizm`, `.prizmkit/config.json` |
+| 1 | Specify | Orchestrator | `prizmkit-specify`, `prizmkit-clarify` | `.prizmkit/specs/spec.md` |
+| 2 | Plan + Tasks | Orchestrator | `prizmkit-plan` | `.prizmkit/specs/plan.md` (含 Tasks section) |
+| 3 | Analyze | Reviewer | `prizmkit-analyze` | Analysis report (no CRITICAL issues) |
+| 4 | Schedule | Orchestrator | — | TaskList entries assigned |
 | 5 | Implement | Dev x N | `prizmkit-implement` | Code + tests, plan.md Tasks marked `[x]` |
-| 6 | Review | QA + Review | `prizmkit-code-review` | Integration tests, review report |
+| 6 | Review | Reviewer | `prizmkit-code-review` | Integration tests, review report |
 | 7 | Fix Loop | Dev | — | Max 3 rounds of fixes |
-| 8 | Summarize & Commit | Coordinator | `prizmkit-summarize`, `prizmkit-committer`, `prizmkit-retrospective` | REGISTRY.md, git commit, .prizm-docs/ updated |
+| 8 | Summarize & Commit | Orchestrator | `prizmkit-summarize`, `prizmkit-committer`, `prizmkit-retrospective` | REGISTRY.md, git commit, .prizm-docs/ updated |
 
 ### Feature Dependency Resolution
 
@@ -461,7 +458,7 @@ The pipeline expects:
 
 | Resource | Location | Description |
 |----------|----------|-------------|
-| Agent Definitions | `.codebuddy/agents/prizm-dev-team-*.md` | 4 agent types: coordinator, pm, dev, reviewer |
+| Agent Definitions | `.codebuddy/agents/prizm-dev-team-*.md` | 2 agent types: dev, reviewer |
 | Team Config | `~/.codebuddy/teams/prizm-dev-team/config.json` | Team runtime configuration |
 | Team Inboxes | `~/.codebuddy/teams/prizm-dev-team/inboxes/` | Agent message inboxes |
 
@@ -469,7 +466,7 @@ The pipeline expects:
 
 | Resource | Location | Description |
 |----------|----------|-------------|
-| Agent Definitions | `.claude/agents/prizm-dev-team-*.md` | 4 agent types: coordinator, pm, dev, reviewer |
+| Agent Definitions | `.claude/agents/prizm-dev-team-*.md` | 2 agent types: dev, reviewer |
 | Team Config | `.claude/team-info.json` | Team runtime configuration (project-level) |
 
 The `generate-bootstrap-prompt.py` script resolves these paths automatically. If paths are incorrect, check the `build_replacements()` function in that script.
@@ -576,7 +573,7 @@ dev-pipeline/bugfix-state/        # Runtime state (gitignored)
 | State dir | `state/` | N/A (in-session) | `bugfix-state/` |
 | Ordering | Dependencies DAG → priority | N/A (single refactor per session) | Severity → priority (no dependencies) |
 | Phases | 10-phase (specify → plan → tasks → implement → review) | 6-phase (analyze → plan → tasks → implement → review → commit) | 5-phase (triage → reproduce → fix → verify → commit) |
-| Agents | Coordinator + PM + Dev + Reviewer | Dev + Reviewer only | Dev + Reviewer only |
+| Agents | Orchestrator + Dev + Reviewer | Dev + Reviewer only | Dev + Reviewer only |
 | Artifacts | spec.md, plan.md, tasks.md, REGISTRY.md | refactor-analysis.md, plan.md, tasks.md | fix-plan.md, fix-report.md only |
 | Commit prefix | `feat(<scope>):` | `refactor(<scope>):` | `fix(<scope>):` |
 | Scope Guard | N/A | ✅ (behavior change → STOP) | N/A |

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

@@ -13,18 +13,14 @@ dev-pipeline (outer loop)
     ├── scripts/                  Python state management scripts
     ├── templates/bootstrap-prompt.md  Session prompt template
     │
-    └── [per session] CodeBuddy CLI
+    └── [per session] AI CLI
             │
-            ├── TeamCreate("prizm-dev-team-F-NNN")
-            ├── Spawn Coordinator agent
-            │       │
-            │       ├── Phase 0: Init
-            │       ├── Phase 1-4: PM (specify/plan/tasks/analyze)
-            │       ├── Phase 5: Schedule & Assign
-            │       ├── Phase 6: Dev x N (parallel implement)
-            │       ├── Phase 7: QA + Review (parallel)
-            │       ├── Phase 8: Fix Loop (max 3 rounds)
-            │       └── Phase 9: Summarize & Commit
+            ├── Phase 0: Init (Orchestrator)
+            ├── Phase 1-2: Context snapshot + Plan & Tasks (Orchestrator)
+            ├── Phase 3-4: Analyze (Reviewer agent)
+            ├── Phase 5: Implement (Dev agent)
+            ├── Phase 6: Review (Reviewer agent)
+            └── Phase 7: Summarize & Commit (Orchestrator)
             │
             └── Write session-status.json → exit
 ```
@@ -33,16 +29,14 @@ dev-pipeline (outer loop)
 
 | Agent | Definition Path | Type |
 |-------|----------------|------|
-| Coordinator | `agent-team-master/prizm-dev-team/prizm-dev-team-coordinator/subagent.md` | prizm-dev-team-coordinator |
-| PM | `agent-team-master/prizm-dev-team/prizm-dev-team-pm/subagent.md` | prizm-dev-team-pm |
-| Dev | `agent-team-master/prizm-dev-team/prizm-dev-team-dev/subagent.md` | prizm-dev-team-dev |
-| QA | `agent-team-master/prizm-dev-team/prizm-dev-team-qa/subagent.md` | prizm-dev-team-qa |
-| Review | `agent-team-master/prizm-dev-team/prizm-dev-team-review/subagent.md` | prizm-dev-team-review |
-| Doc-Reader | `agent-team-master/prizm-dev-team/prizm-dev-team-doc-reader/subagent.md` | prizm-dev-team-doc-reader |
+| Dev | `core/agents/prizm-dev-team-dev.md` | prizm-dev-team-dev |
+| Reviewer | `core/agents/prizm-dev-team-reviewer.md` | prizm-dev-team-reviewer |
+
+Note: The Orchestrator role is handled by the main agent (session orchestrator) directly — no separate agent definition needed.
 
 ## Validator Scripts
 
-Located at `agent-team-master/prizm-dev-team/prizm-dev-team-coordinator/scripts/`:
+Located at `dev-pipeline/scripts/`:
 
 | Script | Checkpoint | Purpose |
 |--------|-----------|---------|
@@ -82,13 +76,13 @@ Located at `agent-team-master/prizm-dev-team/prizm-dev-team-coordinator/scripts/
 ### 1. Session Start
 
 The bootstrap prompt instructs the agent to:
-- Create a team named `prizm-dev-team-{FEATURE_ID}`
-- Spawn the Coordinator with the full subagent.md prompt
-- Coordinator then spawns PM, Dev, QA, Review as needed
+- Execute phases directly as the session orchestrator
+- Spawn Dev and Reviewer agents as subagents for implementation and review phases
+- The orchestrator handles context building, planning, summarize, and commit phases directly
 
 ### 2. Pipeline Execution
 
-The Coordinator drives the 10-phase pipeline with 8 checkpoints (CP-0 through CP-7). Each checkpoint validates artifacts using the validator scripts.
+The Orchestrator drives the pipeline phases with checkpoints (CP-0 through CP-3). Each checkpoint validates artifacts.
 
 ### 3. Session End
 

package/bundled/dev-pipeline/launch-bugfix-daemon.sh

@@ -430,6 +430,14 @@ Examples:
   ./launch-bugfix-daemon.sh logs --follow                 # Live log tailing
   ./launch-bugfix-daemon.sh stop                          # Graceful shutdown
   ./launch-bugfix-daemon.sh restart                       # Stop + start
+
+Environment Variables (pass via --env):
+  MAX_RETRIES           Max retries per bug (default: 3)
+  SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)
+  VERBOSE               Set to 1 for verbose AI CLI output
+  HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)
+  USE_WORKTREE          Enable git worktree isolation per session (default: 1, set 0 to disable)
+  AUTO_PUSH             Auto-push to remote after successful worktree merge (default: 0, set 1 to enable)
 HELP
 }
 

package/bundled/dev-pipeline/launch-daemon.sh

@@ -576,6 +576,8 @@ Environment Variables (pass via --env):
   SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)
   VERBOSE               Set to 1 for verbose AI CLI output
   HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)
+  USE_WORKTREE          Enable git worktree isolation per session (default: 1, set 0 to disable)
+  AUTO_PUSH             Auto-push to remote after successful worktree merge (default: 0, set 1 to enable)
 HELP
 }
 

package/bundled/dev-pipeline/lib/worktree.sh

@@ -0,0 +1,164 @@
+#!/usr/bin/env bash
+# ============================================================
+# dev-pipeline/lib/worktree.sh - Git Worktree Lifecycle Library
+#
+# Shared by run.sh and run-bugfix.sh to isolate AI CLI sessions
+# in separate git worktrees. Each session runs on its own branch
+# inside a worktree directory, enabling parallel-safe execution
+# and clean main branch history.
+#
+# Functions:
+#   worktree_create      — Create worktree + branch before session
+#   worktree_merge       — Merge worktree branch into target after success
+#   worktree_cleanup     — Remove worktree directory and branch
+#   worktree_prune_stale — Prune stale worktree references
+#
+# Environment:
+#   USE_WORKTREE  — Set to 1 to enable (default), 0 to disable
+#   AUTO_PUSH     — Set to 1 to auto-push after successful merge
+# ============================================================
+
+# worktree_create <project_root> <worktree_base_dir> <session_id> <source_branch>
+#
+# Creates a git worktree and sets global vars:
+#   _WORKTREE_PATH   — absolute path to the worktree directory
+#   _WORKTREE_BRANCH — branch name (worktree/<session_id>)
+#
+# Returns 0 on success, 1 on failure.
+worktree_create() {
+    local project_root="$1"
+    local worktree_base_dir="$2"
+    local session_id="$3"
+    local source_branch="$4"
+
+    _WORKTREE_PATH=""
+    _WORKTREE_BRANCH=""
+
+    local branch_name="worktree/${session_id}"
+    local worktree_path="${worktree_base_dir}/${session_id}"
+
+    # Ensure base directory exists
+    mkdir -p "$worktree_base_dir"
+
+    # Check if worktree path already exists
+    if [[ -d "$worktree_path" ]]; then
+        log_warn "Worktree path already exists: $worktree_path"
+        return 1
+    fi
+
+    # Check if branch already exists
+    if git -C "$project_root" rev-parse --verify "$branch_name" >/dev/null 2>&1; then
+        log_warn "Branch already exists: $branch_name"
+        return 1
+    fi
+
+    # Create worktree with new branch
+    if ! git -C "$project_root" worktree add -b "$branch_name" "$worktree_path" "$source_branch" 2>/dev/null; then
+        log_error "Failed to create worktree at $worktree_path"
+        return 1
+    fi
+
+    _WORKTREE_PATH="$worktree_path"
+    _WORKTREE_BRANCH="$branch_name"
+
+    log_info "Created worktree: $worktree_path (branch: $branch_name)"
+    return 0
+}
+
+# worktree_merge <project_root> <worktree_branch> <target_branch> <item_id> <session_id>
+#
+# Merges worktree branch into target branch (no fast-forward).
+# Sets global var:
+#   _MERGE_RESULT — "success", "conflict", or "error"
+#
+# Returns 0 on success, 1 on conflict, 2 on other error.
+worktree_merge() {
+    local project_root="$1"
+    local worktree_branch="$2"
+    local target_branch="$3"
+    local item_id="$4"
+    local session_id="$5"
+
+    _MERGE_RESULT=""
+
+    # Switch to target branch in the main working tree
+    local current_branch
+    current_branch=$(git -C "$project_root" rev-parse --abbrev-ref HEAD 2>/dev/null) || {
+        log_error "Failed to determine current branch"
+        _MERGE_RESULT="error"
+        return 2
+    }
+
+    if [[ "$current_branch" != "$target_branch" ]]; then
+        if ! git -C "$project_root" checkout "$target_branch" 2>/dev/null; then
+            log_error "Failed to checkout target branch: $target_branch"
+            _MERGE_RESULT="error"
+            return 2
+        fi
+    fi
+
+    # Attempt merge (no fast-forward to preserve history)
+    local merge_msg="Merge ${worktree_branch} for ${item_id} (session: ${session_id})"
+    if git -C "$project_root" merge --no-ff -m "$merge_msg" "$worktree_branch" 2>/dev/null; then
+        _MERGE_RESULT="success"
+        log_success "Merged $worktree_branch into $target_branch"
+        return 0
+    else
+        # Check if it's a merge conflict
+        if git -C "$project_root" diff --name-only --diff-filter=U 2>/dev/null | head -1 | read -r _; then
+            # Abort the merge to leave working tree clean
+            git -C "$project_root" merge --abort 2>/dev/null || true
+            _MERGE_RESULT="conflict"
+            log_warn "Merge conflict detected for $worktree_branch"
+            return 1
+        else
+            git -C "$project_root" merge --abort 2>/dev/null || true
+            _MERGE_RESULT="error"
+            log_error "Merge failed for $worktree_branch"
+            return 2
+        fi
+    fi
+}
+
+# worktree_cleanup <project_root> <worktree_path> <worktree_branch>
+#
+# Removes worktree directory and deletes the branch.
+# Idempotent — safe to call even if worktree/branch don't exist.
+#
+# Returns 0 always.
+worktree_cleanup() {
+    local project_root="$1"
+    local worktree_path="$2"
+    local worktree_branch="$3"
+
+    # Remove the worktree (if it exists)
+    if [[ -n "$worktree_path" && -d "$worktree_path" ]]; then
+        git -C "$project_root" worktree remove --force "$worktree_path" 2>/dev/null || {
+            # Fallback: manual removal
+            rm -rf "$worktree_path" 2>/dev/null || true
+        }
+        log_info "Removed worktree: $worktree_path"
+    fi
+
+    # Delete the branch (if it exists)
+    if [[ -n "$worktree_branch" ]]; then
+        git -C "$project_root" branch -D "$worktree_branch" 2>/dev/null || true
+    fi
+
+    # Prune stale worktree entries
+    git -C "$project_root" worktree prune 2>/dev/null || true
+
+    return 0
+}
+
+# worktree_prune_stale <project_root>
+#
+# Runs `git worktree prune` to clean up stale worktree references.
+#
+# Returns 0 always.
+worktree_prune_stale() {
+    local project_root="$1"
+
+    git -C "$project_root" worktree prune 2>/dev/null || true
+    return 0
+}

package/bundled/dev-pipeline/retry-bug.sh

@@ -232,17 +232,20 @@ if [[ -n "${MODEL:-}" ]]; then
     MODEL_FLAG="--model $MODEL"
 fi
 
+unset CLAUDECODE 2>/dev/null || true
+
 case "$CLI_CMD" in
     *claude*)
+        # Claude Code: prompt via -p argument, --dangerously-skip-permissions for auto-accept
         "$CLI_CMD" \
-            --print \
             -p "$(cat "$BOOTSTRAP_PROMPT")" \
-            --yes \
+            --dangerously-skip-permissions \
             $STREAM_JSON_FLAG \
             $MODEL_FLAG \
             > "$SESSION_LOG" 2>&1 &
         ;;
     *)
+        # CodeBuddy (cbc) and others: prompt via stdin
         "$CLI_CMD" \
             --print \
             -y \

package/bundled/dev-pipeline/retry-feature.sh

@@ -233,17 +233,20 @@ if [[ -n "${MODEL:-}" ]]; then
     MODEL_FLAG="--model $MODEL"
 fi
 
+unset CLAUDECODE 2>/dev/null || true
+
 case "$CLI_CMD" in
     *claude*)
+        # Claude Code: prompt via -p argument, --dangerously-skip-permissions for auto-accept
         "$CLI_CMD" \
-            --print \
             -p "$(cat "$BOOTSTRAP_PROMPT")" \
-            --yes \
+            --dangerously-skip-permissions \
             $STREAM_JSON_FLAG \
             $MODEL_FLAG \
             > "$SESSION_LOG" 2>&1 &
         ;;
     *)
+        # CodeBuddy (cbc) and others: prompt via stdin
         "$CLI_CMD" \
             --print \
             -y \