@xulthekl/team-flow 0.34.1 → 0.36.0
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.
- package/.claude/always/phase-guard.md +1 -1
- package/.claude-plugin/marketplace.json +1 -1
- package/.claude-plugin/plugin.json +1 -1
- package/.codex-plugin/plugin.json +1 -1
- package/.cursor-plugin/marketplace.json +1 -1
- package/.cursor-plugin/plugin.json +1 -1
- package/.github/plugin/marketplace.json +2 -2
- package/CHANGELOG.md +57 -0
- package/GEMINI.md +1 -1
- package/INSTALL.md +1 -1
- package/README.md +1 -1
- package/agents/architecture-reviewer.md +12 -0
- package/agents/build-executor.md +23 -0
- package/docs/README_en.md +1 -1
- package/docs/solutions/INDEX.md +1 -0
- package/docs/solutions/cross-phase/2026-08-05-no-summary.md +17 -0
- package/gemini-extension.json +1 -1
- package/hooks/session-start +2 -2
- package/llms.txt +1 -1
- package/package.json +1 -1
- package/plugin.json +1 -1
- package/scripts/guard/checks/arch-gate-exemptions.mjs +68 -0
- package/scripts/guard/checks/arch-readiness.mjs +36 -0
- package/scripts/guard/checks/arch-snapshot.mjs +35 -0
- package/scripts/guard/guard.mjs +10 -2
- package/scripts/lib/arch-merge.mjs +405 -316
- package/scripts/lib/arch-parse.mjs +162 -0
- package/scripts/lib/cmd-arch.mjs +84 -0
- package/scripts/lib/cmd-deisolate.mjs +292 -0
- package/scripts/lib/cmd-execution.mjs +16 -3
- package/scripts/lib/execution-plan.mjs +37 -5
- package/scripts/lib/test-merge.mjs +7 -4
- package/scripts/team-flow.mjs +7 -0
- package/skills/architecture-design/SKILL.md +22 -1
- package/skills/architecture-design/chapters/ch06-integration.md +4 -4
- package/skills/architecture-design/references/s3.5-architecture-template.md +164 -0
- package/skills/architecture-design/references/s3.5-loading-protocol.md +40 -0
- package/skills/architecture-design/references/s3.5-product-architecture.md +76 -0
- package/skills/release-archivist/SKILL.md +14 -0
- package/skills/session-handoff/references/handoff-template.md +2 -2
- package/skills/spec-writer/SKILL.md +1 -0
- package/skills/workflow-bootstrap/references/agents/arch-reverse-analyst.md +60 -0
- package/skills/workflow-orchestrator/references/feedback-loops.md +2 -0
- package/skills/workflow-orchestrator/references/s1-path-router.md +1 -1
- package/skills/workflow-orchestrator/references/s3-plan-pipeline.md +2 -1
- package/skills/workflow-orchestrator/references/s4-split-validate.md +11 -1
- package/skills/workflow-orchestrator/references/state-model.md +52 -0
- package/skills/workflow-start/SKILL.md +1 -0
- package/skills/workflow-start/references/routing-rules.md +1 -0
|
@@ -61,6 +61,7 @@ Validate mode against artifact content. If hotfix/tweak criteria not met → upg
|
|
|
61
61
|
产出型路由(spec-writer / contract-builder / build-executor / release-archivist)通过 `Agent(subagent_type: "team-flow:<name>")` **dispatch 子代理**执行——agent 定义经 `skills:` 预加载对应 SKILL.md,主代理**不得**凭记忆手写产物格式 prompt。need-explorer 为交互式澄清,**主进程执行**(需与用户对话;其 agent 定义 `team-flow:need-explorer` 仅作可 dispatch 后备)。
|
|
62
62
|
|
|
63
63
|
- **记录 agentId**:dispatch 后立即记录子代理 ID;修复一律 SendMessage 恢复原子代理(禁启动新子代理,见 v0.29.0 §37)
|
|
64
|
+
- **上下文水位例外(v0.35.0)**:当原子代理 transcript 超过 2MB 或累计 token 消耗超过 200 万时,主代理可以启动新代理替代恢复。新代理必须接收完整交接文档(当前进度、已完成 wave 列表、未解决问题、artifact 位置、已知 blocker),以保证连续性。此为 SendMessage 恢复原则的唯一补充例外
|
|
64
65
|
- **返回即验证(validation gate)**:子代理返回产物后**立即** `tf validate <change-dir>`;FAIL → SendMessage 回**原**子代理修复,通过后才能继续/转换。把格式失败拦截在返回时,而非状态转换时(避免浪费整次执行后再失败)。**build-executor 附加验证(v0.13 §52 B3,SDD/full 模式)**:返回的 diff 中必须包含测试文件(src/test/ 或项目测试目录);零测试文件的实现返回一律 BLOCK,SendMessage 回原子代理按 test-matrix.md / TDD Iron Law 补齐——不得以"手动验证"替代(C1-domain-policy 事件教训)。例外:已显式记录 `test_matrix_skipped=true` + 理由的 change(纯文档/配置类)不要求测试文件
|
|
65
66
|
- **结果协议**:子代理终态须标注 `FINAL VERDICT: <DONE|BLOCKED|FAIL>`(审查类用 PASS/PASS_WITH_WARNINGS/FAIL);**SendMessage 报告为权威结果**,task-notification.result 仅内部元数据
|
|
66
67
|
- **子代理状态边界**:子代理 MUST NOT 修改 `.team-flow.yaml` 的 `state`/`workflow` 核心字段(状态转换是主代理专有职责,经 `tf state transition` 执行),只写自己的 `dp_N_*` 决策字段
|
|
@@ -69,6 +69,7 @@ reviewer 报告 → PASS / PASS_WITH_WARNINGS → 进入 Step 3
|
|
|
69
69
|
- 重新审查:`SendMessage(to: arch_reviewer_agent_id, message: "第 {N} 轮审查,架构产物已修正。请重新执行 6 维度审查。")`
|
|
70
70
|
- **禁止**在循环中启动新的 architecture-design 或 architecture-reviewer 子代理——原子代理拥有完整的设计上下文和审查历史,新子代理需要重新加载全部上下文(浪费 10-20K tokens 且可能丢失修正连贯性)
|
|
71
71
|
- **唯一例外**:SendMessage 恢复失败(子代理 transcript 不可用)时,启动新子代理作为 fallback,但必须在 dispatch prompt 中传入:原产物路径 + auto-review 报告 + 历次修正记录
|
|
72
|
+
- **上下文水位例外(v0.35.0)**:当原子代理 transcript 超过 2MB 或累计 token 消耗超过 200 万时,主代理可以启动新代理替代恢复。新代理必须接收完整交接文档(当前进度、已完成 wave 列表、未解决问题、artifact 位置、已知 blocker),以保证连续性
|
|
72
73
|
|
|
73
74
|
报告落盘到 `changes/<name>/architecture/auto-review.md`。
|
|
74
75
|
|