harnessed 1.0.4 → 2.0.1

package/dist/index.mjs

@@ -1,6 +1,6 @@
 // package.json
 var package_default = {
-  version: "1.0.4"};
+  version: "2.0.1"};
 
 // src/index.ts
 var VERSION = package_default.version;

package/dist/index.mjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../package.json","../src/index.ts"],"names":[],"mappings":";AAAA,IAAA,eAAA,GAAA;AAAA,EAEE,OAAA,EAAW,OA6Fb,CAAA;;;AC1FO,IAAM,UAAU,eAAA,CAAI","file":"index.mjs","sourcesContent":["{\n  \"name\": \"harnessed\",\n  \"version\": \"1.0.4\",\n  \"description\": \"AI coding harness package manager + composition orchestrator\",\n  \"type\": \"module\",\n  \"license\": \"Apache-2.0\",\n  \"author\": \"easyinplay\",\n  \"repository\": {\n    \"type\": \"git\",\n    \"url\": \"https://github.com/easyinplay/harnessed.git\"\n  },\n  \"homepage\": \"https://github.com/easyinplay/harnessed#readme\",\n  \"bugs\": \"https://github.com/easyinplay/harnessed/issues\",\n  \"keywords\": [\n    \"claude-code\",\n    \"ai-harness\",\n    \"package-manager\",\n    \"composition\",\n    \"skill-pack\",\n    \"mcp\",\n    \"orchestrator\"\n  ],\n  \"engines\": {\n    \"node\": \">=22.0.0\"\n  },\n  \"packageManager\": \"pnpm@10.12.0\",\n  \"bin\": {\n    \"harnessed\": \"./dist/cli.mjs\"\n  },\n  \"main\": \"./dist/index.mjs\",\n  \"types\": \"./dist/index.d.ts\",\n  \"exports\": {\n    \".\": {\n      \"types\": \"./dist/index.d.ts\",\n      \"import\": \"./dist/index.mjs\",\n      \"default\": \"./dist/index.mjs\"\n    },\n    \"./schemas\": {\n      \"types\": \"./dist/schemas/index.d.ts\",\n      \"import\": \"./dist/schemas/index.mjs\"\n    },\n    \"./package.json\": \"./package.json\"\n  },\n  \"files\": [\n    \"dist\",\n    \"manifests\",\n    \"workflows\",\n    \"routing\",\n    \"config-templates\",\n    \"schemas\",\n    \"README.md\",\n    \"LICENSE\",\n    \"NOTICE\"\n  ],\n  \"scripts\": {\n    \"dev\": \"tsup --watch\",\n    \"build\": \"tsc --noEmit && tsup\",\n    \"build:schema\": \"node ./scripts/build-schema.mjs\",\n    \"validate:schema\": \"node ./scripts/validate-schema.mjs\",\n    \"typecheck\": \"tsc --noEmit\",\n    \"test\": \"vitest run --passWithNoTests\",\n    \"test:watch\": \"vitest\",\n    \"test:coverage\": \"vitest run --coverage --passWithNoTests\",\n    \"bench\": \"vitest bench --run\",\n    \"lint\": \"biome check .\",\n    \"lint:fix\": \"biome check --write .\",\n    \"format\": \"biome format --write .\"\n  },\n  \"dependencies\": {\n    \"@anthropic-ai/claude-agent-sdk\": \"0.3.142\",\n    \"@clack/prompts\": \"^0.10.1\",\n    \"@sinclair/typebox\": \"^0.34.49\",\n    \"ajv\": \"^8.20.0\",\n    \"ajv-errors\": \"^3.0.0\",\n    \"ajv-formats\": \"^3.0.1\",\n    \"commander\": \"^13.0.0\",\n    \"diff\": \"^9.0.0\",\n    \"picocolors\": \"^1.1.1\",\n    \"proper-lockfile\": \"^4.1.2\",\n    \"yaml\": \"^2.9.0\"\n  },\n  \"devDependencies\": {\n    \"@biomejs/biome\": \"^2.0.0\",\n    \"@types/node\": \"^22.10.0\",\n    \"@types/proper-lockfile\": \"^4.1.4\",\n    \"@vitest/coverage-v8\": \"^4.0.0\",\n    \"tsup\": \"^8.3.0\",\n    \"typescript\": \"^5.6.0\",\n    \"vitest\": \"^4.0.0\"\n  },\n  \"pnpm\": {\n    \"onlyBuiltDependencies\": [\n      \"esbuild\"\n    ]\n  }\n}\n","// Main library entry — re-exports public APIs.\n// phase 1.1 batch 1: skeleton only; schema validator wired in batch 2 (T3+).\n\nimport pkg from '../package.json' with { type: 'json' }\n\nexport const VERSION = pkg.version\n"]}
+{"version":3,"sources":["../package.json","../src/index.ts"],"names":[],"mappings":";AAAA,IAAA,eAAA,GAAA;AAAA,EAEE,OAAA,EAAW,OA8Fb,CAAA;;;AC3FO,IAAM,UAAU,eAAA,CAAI","file":"index.mjs","sourcesContent":["{\n  \"name\": \"harnessed\",\n  \"version\": \"2.0.1\",\n  \"description\": \"AI coding harness package manager + composition orchestrator\",\n  \"type\": \"module\",\n  \"license\": \"Apache-2.0\",\n  \"author\": \"easyinplay\",\n  \"repository\": {\n    \"type\": \"git\",\n    \"url\": \"https://github.com/easyinplay/harnessed.git\"\n  },\n  \"homepage\": \"https://github.com/easyinplay/harnessed#readme\",\n  \"bugs\": \"https://github.com/easyinplay/harnessed/issues\",\n  \"keywords\": [\n    \"claude-code\",\n    \"ai-harness\",\n    \"package-manager\",\n    \"composition\",\n    \"skill-pack\",\n    \"mcp\",\n    \"orchestrator\"\n  ],\n  \"engines\": {\n    \"node\": \">=22.0.0\"\n  },\n  \"packageManager\": \"pnpm@10.12.0\",\n  \"bin\": {\n    \"harnessed\": \"./dist/cli.mjs\"\n  },\n  \"main\": \"./dist/index.mjs\",\n  \"types\": \"./dist/index.d.ts\",\n  \"exports\": {\n    \".\": {\n      \"types\": \"./dist/index.d.ts\",\n      \"import\": \"./dist/index.mjs\",\n      \"default\": \"./dist/index.mjs\"\n    },\n    \"./schemas\": {\n      \"types\": \"./dist/schemas/index.d.ts\",\n      \"import\": \"./dist/schemas/index.mjs\"\n    },\n    \"./package.json\": \"./package.json\"\n  },\n  \"files\": [\n    \"dist\",\n    \"manifests\",\n    \"workflows\",\n    \"routing\",\n    \"config-templates\",\n    \"schemas\",\n    \"README.md\",\n    \"LICENSE\",\n    \"NOTICE\"\n  ],\n  \"scripts\": {\n    \"dev\": \"tsup --watch\",\n    \"build\": \"tsc --noEmit && tsup\",\n    \"build:schema\": \"node ./scripts/build-schema.mjs\",\n    \"validate:schema\": \"node ./scripts/validate-schema.mjs\",\n    \"typecheck\": \"tsc --noEmit\",\n    \"test\": \"vitest run --passWithNoTests\",\n    \"test:watch\": \"vitest\",\n    \"test:coverage\": \"vitest run --coverage --passWithNoTests\",\n    \"bench\": \"vitest bench --run\",\n    \"lint\": \"biome check .\",\n    \"lint:fix\": \"biome check --write .\",\n    \"format\": \"biome format --write .\"\n  },\n  \"dependencies\": {\n    \"@anthropic-ai/claude-agent-sdk\": \"0.3.142\",\n    \"@clack/prompts\": \"^0.10.1\",\n    \"@sinclair/typebox\": \"^0.34.49\",\n    \"ajv\": \"^8.20.0\",\n    \"ajv-errors\": \"^3.0.0\",\n    \"ajv-formats\": \"^3.0.1\",\n    \"commander\": \"^13.0.0\",\n    \"diff\": \"^9.0.0\",\n    \"expr-eval\": \"^2.0.2\",\n    \"picocolors\": \"^1.1.1\",\n    \"proper-lockfile\": \"^4.1.2\",\n    \"yaml\": \"^2.9.0\"\n  },\n  \"devDependencies\": {\n    \"@biomejs/biome\": \"^2.0.0\",\n    \"@types/node\": \"^22.10.0\",\n    \"@types/proper-lockfile\": \"^4.1.4\",\n    \"@vitest/coverage-v8\": \"^4.0.0\",\n    \"tsup\": \"^8.3.0\",\n    \"typescript\": \"^5.6.0\",\n    \"vitest\": \"^4.0.0\"\n  },\n  \"pnpm\": {\n    \"onlyBuiltDependencies\": [\n      \"esbuild\"\n    ]\n  }\n}\n","// Main library entry — re-exports public APIs.\n// phase 1.1 batch 1: skeleton only; schema validator wired in batch 2 (T3+).\n\nimport pkg from '../package.json' with { type: 'json' }\n\nexport const VERSION = pkg.version\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "harnessed",
-  "version": "1.0.4",
+  "version": "2.0.1",
   "description": "AI coding harness package manager + composition orchestrator",
   "type": "module",
   "license": "Apache-2.0",
@@ -75,6 +75,7 @@
     "ajv-formats": "^3.0.1",
     "commander": "^13.0.0",
     "diff": "^9.0.0",
+    "expr-eval": "^2.0.2",
     "picocolors": "^1.1.1",
     "proper-lockfile": "^4.1.2",
     "yaml": "^2.9.0"

package/workflows/capabilities.yaml

@@ -0,0 +1,413 @@
+# <packageRoot>/workflows/capabilities.yaml — harnessed v2.0 capabilities baseline
+#
+# Maintainer-curated static manifest mapping abstract capability name → concrete
+# impl + slash-cmd. Consumed by workflows/*/workflow.yaml v2 via `{{ capabilities.<name>.cmd }}`
+# template interpolation (per R20.2 flat yaml map + R20.5 static manifest + ADR per upgrade).
+#
+# Pure bundled distribution (per R20.1 + D-01): end-user 不 override; maintainer 改此文件 +
+# ADR 0024+ + npm patch release (2-3 day cycle). Schema = harnessed.capabilities.v1 (sister
+# Phase 2.2 schemaVersion 7 surface convention).
+#
+# Entry buckets (5, total 35):
+#   1. mattpocock 12 高频招式 (D-09)        — grill-with-docs / zoom-out / diagnose / caveman /
+#                                              grill-me / tdd / to-prd / to-issues /
+#                                              improve-codebase-architecture / code-review /
+#                                              code-simplifier / investigate
+#   2. special-purpose tools 13+ (D-14)     — ui-ux-pro-max / frontend-design / playwright-cli /
+#                                              playwright-test / webapp-testing /
+#                                              chrome-devtools-mcp / ctx7 / tavily-mcp /
+#                                              exa-mcp / gsd-review / gsd-debug /
+#                                              gsd-progress / gsd-verify-work
+#   3. gstack 治理关卡 6 (D-12)              — gstack-office-hours / gstack-plan-ceo-review /
+#                                              gstack-review / gstack-qa / gstack-cso /
+#                                              gstack-design-review
+#   4. 核心 capability 4                    — tdd (D-13 alias) / planning-with-files (D-15 +
+#                                              Q-AUDIT-5a claude-code-plugin reframe) /
+#                                              ralph-loop (D-10) / superpowers-brainstorming
+#   5. Agent Teams 3 (D-11 + Q-AUDIT-5b)    — agent-teams-create / agent-teams-send-message /
+#                                              agent-teams-shutdown
+schema_version: harnessed.capabilities.v1
+
+capabilities:
+
+  # ============================================================================
+  # Bucket 1 — mattpocock 12 高频招式 (per D-09 + RESEARCH § 2.1)
+  # ============================================================================
+
+  grill-with-docs:
+    impl: mattpocock-skills
+    cmd: /grill-with-docs
+    since: v2.0
+    description: 规格澄清 with docs (fires when phase.spec_ambiguous == true)
+    fires_when:
+      - phase.spec_ambiguous == true
+
+  zoom-out:
+    impl: mattpocock-skills
+    cmd: /zoom-out
+    since: v2.0
+    description: 陌生模块导航 (fires when subtask.unfamiliar_module == true)
+    fires_when:
+      - subtask.unfamiliar_module == true
+
+  diagnose:
+    impl: mattpocock-skills
+    cmd: /diagnose
+    since: v2.0
+    description: 系统化排错 (fires when test_fail OR debug context)
+    fires_when:
+      - subtask.test_fail == true
+      - subtask.type == 'debug'
+
+  caveman:
+    impl: mattpocock-skills
+    cmd: /caveman
+    since: v2.0
+    description: token 紧张时 surgical edit (fires when token_tight == true)
+    fires_when:
+      - subtask.token_tight == true
+
+  grill-me:
+    impl: mattpocock-skills
+    cmd: /grill-me
+    since: v2.0
+    description: 反向追问 (fires when spec_ambiguous AND no_docs)
+    fires_when:
+      - phase.spec_ambiguous == true AND phase.no_docs == true
+
+  to-prd:
+    impl: mattpocock-skills
+    cmd: /to-prd
+    since: v2.0
+    description: translate idea → PRD (fires when phase.translate_idea_to_prd == true)
+    fires_when:
+      - phase.translate_idea_to_prd == true
+
+  to-issues:
+    impl: mattpocock-skills
+    cmd: /to-issues
+    since: v2.0
+    description: translate PRD → issues (fires when phase.translate_prd_to_issues == true)
+    fires_when:
+      - phase.translate_prd_to_issues == true
+
+  improve-codebase-architecture:
+    impl: mattpocock-skills
+    cmd: /improve-codebase-architecture
+    since: v2.0
+    description: architecture health audit (fires when phase.architecture_health_audit == true)
+    fires_when:
+      - phase.architecture_health_audit == true
+
+  code-review:
+    impl: mattpocock-skills
+    cmd: /code-review
+    since: v2.0
+    description: verify-work parallel fan-out (multi-agent high-confidence findings)
+    fires_when:
+      - phase.stage == 'verify'
+
+  code-simplifier:
+    impl: mattpocock-skills
+    cmd: /code-simplifier
+    since: v2.0
+    description: verify-work tail step (移除重复、多余逻辑)
+    fires_when:
+      - phase.stage == 'verify' AND phase.is_final_step == true
+
+  investigate:
+    impl: mattpocock-skills
+    cmd: /investigate
+    since: v2.0
+    description: gstack-routed bug investigation (fires when bug_root_cause_unknown)
+    fires_when:
+      - subtask.bug_root_cause_unknown == true
+
+  # Note: 剩 11 mattpocock 招式 (e.g., /retro, /ship, /office-hours-alias 等) 推 v2.x patch
+  # release per D-09 (manifest-only since: v2.x — 实际未 wire 至 workflow.yaml 的 entry 不在
+  # v2.0 baseline 中)。
+
+  # ============================================================================
+  # Bucket 2 — special-purpose tools 13 (per D-14 + RESEARCH § 2.2)
+  # ============================================================================
+
+  ui-ux-pro-max:
+    impl: gstack
+    cmd: /ui-ux-pro-max
+    since: v2.0
+    description: UI/UX 默认主方案 数据驱动 (per ~/.claude/rules/web-design.md L5)
+    fires_when:
+      - phase.has_ui_changes == true
+      - subtask.type == 'ui_polish'
+
+  frontend-design:
+    impl: gstack
+    cmd: /frontend-design
+    since: v2.0
+    description: UI 创意/装饰补充 (per ~/.claude/rules/web-design.md L11)
+    fires_when:
+      - phase.has_ui_changes == true AND phase.requires_creative_polish == true
+
+  playwright-cli:
+    impl: npm-cli
+    cmd: playwright
+    since: v2.0
+    description: AI 实时操作浏览器 一行 Bash (per ~/.claude/rules/web-testing.md L8 手)
+    fires_when:
+      - subtask.type == 'browser_probe'
+
+  playwright-test:
+    impl: npm-cli
+    cmd: "@playwright/test"
+    since: v2.0
+    description: E2E test framework TypeScript (per ~/.claude/rules/web-testing.md L8 脑)
+    fires_when:
+      - subtask.type == 'e2e_test' AND subtask.language == 'typescript'
+
+  webapp-testing:
+    impl: gstack
+    cmd: /webapp-testing
+    since: v2.0
+    description: Python 后端联动 E2E (per ~/.claude/rules/web-testing.md L11 筋骨)
+    fires_when:
+      - subtask.type == 'e2e_test' AND subtask.language == 'python'
+
+  chrome-devtools-mcp:
+    impl: mcp
+    cmd: chrome-devtools
+    since: v2.0
+    description: 性能/a11y/内存泄漏诊断 (per ~/.claude/rules/web-testing.md L45 非功能性)
+    fires_when:
+      - subtask.type in ['perf_diagnose', 'a11y_audit', 'memory_leak']
+
+  ctx7:
+    impl: cli
+    cmd: ctx7
+    since: v2.0
+    description: 库 API/框架/SDK 文档抓取 (per ~/.claude/rules/context7.md)
+    fires_when:
+      - subtask.needs_library_docs == true
+
+  tavily-mcp:
+    impl: mcp
+    cmd: tavily_search
+    since: v2.0
+    description: Web 搜索默认 (关键词/库 API/时效) per ~/.claude/rules/web-search.md L17
+    fires_when:
+      - subtask.needs_web_search == true
+
+  exa-mcp:
+    impl: mcp
+    cmd: web_fetch_exa
+    since: v2.0
+    description: 描述式/学术 Web 搜索 per ~/.claude/rules/web-search.md L24
+    fires_when:
+      - subtask.needs_web_search == true AND subtask.search_type in ['descriptive', 'academic']
+
+  gsd-discuss-phase:
+    impl: gsd
+    cmd: /gsd-discuss-phase
+    since: v2.0
+    description: GSD discuss-phase 灰色澄清 (Phase 层判据 sister judgments/phase-gate.yaml)
+    fires_when:
+      - phase.open_decisions >= 2 OR phase.scope_days > 1
+
+  gsd-plan-phase:
+    impl: gsd
+    cmd: /gsd-plan-phase
+    since: v2.0
+    description: GSD plan-phase Wave A research + Wave B planner + Wave C plan-checker
+    fires_when:
+      - phase.stage == 'plan'
+
+  gsd-review:
+    impl: gsd
+    cmd: /gsd-review
+    since: v2.0
+    description: GSD cross-AI peer review
+    fires_when:
+      - phase.stage == 'verify' AND phase.requires_peer_review == true
+
+  gsd-debug:
+    impl: gsd
+    cmd: /gsd-debug
+    since: v2.0
+    description: GSD systematic debugging (overlap with mattpocock /diagnose — alias)
+    aliases:
+      - { impl: mattpocock-skills, cmd: /diagnose }
+    fires_when:
+      - subtask.type == 'debug'
+
+  gsd-progress:
+    impl: gsd
+    cmd: /gsd-progress
+    since: v2.0
+    description: GSD 状态同步 (verify-work serial after gsd-verify-work)
+    fires_when:
+      - phase.stage == 'verify'
+
+  gsd-verify-work:
+    impl: gsd
+    cmd: /gsd-verify-work
+    since: v2.0
+    description: verify-work entry (Stage ④ Verify 起点)
+    fires_when:
+      - phase.stage == 'verify'
+
+  # ============================================================================
+  # Bucket 3 — gstack 治理关卡 6 (per D-12 + CLAUDE.md "gstack 治理关卡")
+  # ============================================================================
+
+  gstack-office-hours:
+    impl: gstack
+    cmd: /office-hours
+    since: v2.0
+    description: 战略层产品理念/商业 scope 澄清 (sister CLAUDE.md "新功能启动前强制")
+    fires_when:
+      - phase.type in ['new_project', 'new_milestone', 'new_feature']
+
+  gstack-plan-ceo-review:
+    impl: gstack
+    cmd: /plan-ceo-review
+    since: v2.0
+    description: 战略评估 CEO 视角 (新功能启动前强制 sister CLAUDE.md)
+    fires_when:
+      - phase.type == 'new_feature' AND phase.has_business_decisions == true
+
+  gstack-review:
+    impl: gstack
+    cmd: /review
+    since: v2.0
+    description: Paranoid Staff Engineer 视角 (关键模块 PR 前强制 sister CLAUDE.md)
+    fires_when:
+      - phase.is_critical_module == true
+
+  gstack-qa:
+    impl: gstack
+    cmd: /qa
+    since: v2.0
+    description: 端到端 QA 验收 (可选 conditional)
+    fires_when:
+      - phase.stage == 'verify' AND phase.has_ui_changes == true
+
+  gstack-cso:
+    impl: gstack
+    cmd: /cso
+    since: v2.0
+    description: 安全审查 OWASP/auth/secrets (可选 conditional)
+    fires_when:
+      - phase.stage == 'verify' AND phase.has_auth_or_secrets == true
+
+  gstack-design-review:
+    impl: gstack
+    cmd: /design-review
+    since: v2.0
+    description: 设计系统一致性审查 + AI 审美问题识别 (可选 conditional)
+    fires_when:
+      - phase.stage == 'verify' AND phase.has_design_changes == true
+
+  # ============================================================================
+  # Bucket 4 — 核心 capability 4 (per D-10 + D-13 + D-15 Q-AUDIT-5a + RESEARCH § 2.3)
+  # ============================================================================
+
+  # D-13: tdd entry 2 impl 候选 alias (superpowers 主 + mattpocock /tdd 备)
+  tdd:
+    impl: superpowers
+    cmd: "superpowers:test-driven-development"
+    since: v2.0
+    description: TDD red-green-refactor (gated by judgments/tdd-gate.yaml fires_when)
+    aliases:
+      - { impl: mattpocock-skills, cmd: /tdd }
+    fires_when:
+      - subtask.is_core_business_logic == true
+      - subtask.is_algorithm == true
+      - subtask.is_data_processing == true
+      - subtask.regression_risk == 'high'
+      - subtask.reliability_required == true
+
+  # D-15 + Q-AUDIT-5a: planning-with-files impl = claude-code-plugin (NOT npm-sdk; verified
+  # 2026-05-20 npm registry 404 reframe terminology drift). Real Claude Code plugin slash
+  # cmd `/plan` invocation → plugin 真生成 3 file 持久化, NOT fs.writeFile self-impl。
+  planning-with-files:
+    impl: claude-code-plugin
+    cmd: /plan
+    since: v2.0
+    description: Manus-style persistent markdown planning (Claude Code plugin slash cmd)
+    requires:
+      plugin: planning-with-files >=2.2.0
+    plugin_path: ~/.claude/plugins/cache/planning-with-files/planning-with-files/2.34.0
+    outputs:
+      - task_plan.md
+      - progress.md
+      - findings.md
+    fires_when:
+      - phase.stage == 'plan' AND phase.requires_persisted_plan == true
+
+  # D-10: ralph-loop 真接 SDK wrapper (NOT mock reference; sister Phase 2.2 v0.2.0 ship)
+  ralph-loop:
+    impl: bundled-skill
+    cmd: ralph-loop
+    since: v2.0
+    description: Sub-task completion gate (verbatim COMPLETE; max-iterations fallback)
+    sdk_ref: src/routing/lib/ralphLoop.ts
+    fires_when:
+      - subtask.completion_required == true
+
+  superpowers-brainstorming:
+    impl: superpowers
+    cmd: "superpowers:brainstorming"
+    since: v2.0
+    description: 子任务级澄清 multi-approach 探索 (gated by judgments/subtask-gate.yaml)
+    fires_when:
+      - subtask.approaches >= 2
+      - subtask.core_algorithm == true
+      - subtask.has_api_contract == true
+      - subtask.error_cost == 'high'
+
+  # ============================================================================
+  # Bucket 5 — Agent Teams 3 升级触发 (per D-11 + Q-AUDIT-5b schema fix +
+  #            ~/.claude/rules/agent-teams.md L7 prereq)
+  # ============================================================================
+  #
+  # Q-AUDIT-5b 2026-05-20 fix: settings_env_var root-level `env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`
+  # (NOT nested experimental.* — Wave A 本地 settings.json 实证修正 prior schema 错误)。
+  # Per Q-AUDIT-2 D-11: harnessed doctor + setup 检查 env var present + cc_version ≥2.1.133。
+  #
+  # 5 fires_when OR-chain (multi-select per ~/.claude/rules/agent-teams.md 升级触发):
+  #   1. teammate_send_message_needed — teammate 间需 SendMessage 互通 (NOT fire-and-forget)
+  #   2. subagent_context_overflow    — 某 subagent 撞 context 上限需分担
+  #   3. shared_task_list             — 多 teammate 共享 task list 自协调
+  #   4. opposing_hypothesis_debate   — 对立假设辩论 (两 teammate 各守一假说 + lead 裁判)
+  #   5. fullstack_three_way          — 全栈三路协同 (前端 + 后端 + 测试同步 + API contract)
+  agent-teams-create:
+    impl: claude-platform
+    cmd: TeamCreate
+    since: v2.0
+    description: Agent Teams team 创建 (CC 2.1.133+ experimental; per agent-teams.md L7)
+    requires:
+      settings_env_var: env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS == "1"
+      cc_version: ">=2.1.133"
+    fires_when:
+      - teammate_send_message_needed == true
+      - subagent_context_overflow == true
+      - shared_task_list == true
+      - opposing_hypothesis_debate == true
+      - fullstack_three_way == true
+
+  agent-teams-send-message:
+    impl: claude-platform
+    cmd: SendMessage
+    since: v2.0
+    description: Agent Teams teammate 间通信 (round-trip 已验证 2026-05-19)
+    requires:
+      settings_env_var: env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS == "1"
+      cc_version: ">=2.1.133"
+
+  agent-teams-shutdown:
+    impl: claude-platform
+    cmd: TeamDelete
+    since: v2.0
+    description: Agent Teams cleanup (shutdown_request 之后 TeamDelete; sister agent-teams.md 防呆清单)
+    requires:
+      settings_env_var: env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS == "1"
+      cc_version: ">=2.1.133"

package/workflows/defaults.yaml

@@ -0,0 +1,36 @@
+# harnessed v2.0 workflow defaults
+# Phase v2.0-2.3 W1.2 ship — ralph_max_iterations table per R20.10 + D-10 fallback abort path
+# Sister Phase 2.2 STRIDE T-2.2-05 DoS mitigation: hard_upper_limit = 100 caps any user override
+# Phase 2.4 workflow.yaml v2 interpolation: {{ defaults.ralph_max_iterations.<workflow>.<phase> }}
+
+schema_version: harnessed.defaults.v1
+
+# Per-workflow per-phase ralph-loop iteration ceiling
+# Values balance: completion guarantee vs DoS prevention; sister ADR 0011 v0.2.0 ralph-loop SDK pattern
+ralph_max_iterations:
+  execute-task:
+    01-clarify: 5
+    02-code: 20
+    03-test: 15
+    04-deliver: 20
+
+  plan-feature:
+    01-gstack-decision: 1
+    02-brainstorm: 5
+    03-gsd-discuss: 3
+    04-gsd-plan: 3
+    05-persist: 5
+
+  research:
+    01-fan-out: 3
+    02-synth: 3
+
+  verify-work:
+    01-gsd-verify-work: 3
+    02-code-review: 5
+    03-gstack-review: 3
+    04-code-simplifier: 5
+
+# Hard upper limit — workflow engine rejects any user override > this value
+# Sister Phase 2.2 STRIDE T-2.2-05 DoS mitigation
+hard_upper_limit: 100

package/workflows/execute-task/SKILL.md

@@ -1,8 +1,11 @@
 ---
 name: execute-task
 description: |
-  execute-task workflow — 4-phase chain (brainstorming → karpathy → mattpocock → TDD → ralph-loop)
+  execute-task workflow v2 — 4-phase chain (brainstorming → karpathy + mattpocock route → TDD + diagnose → ralph-loop COMPLETE)
   triggered by harnessed CLI `harnessed execute-task --task <text>`.
+  v2 delta (Phase v2.0-2.4 W1.1): schema_version: harnessed.workflow.v2 + ralph-loop 真接 SDK wrapper +
+  tdd-gate conditional + mattpocock route by condition (grill-with-docs / zoom-out / diagnose) +
+  explicit max_iterations_exceeded fallback (R20.10 NOT silent abort).
 trigger_phrases:
   # forward-looking documentation — auto-invocation 实装推 Phase 2.3 extension category (B-28).
   # Current enforced entry surface is the CLI subcommand below; these phrases are
@@ -13,23 +16,33 @@ trigger_phrases:
   - "跑 execute-task"
 ---
 
-# execute-task workflow
+# execute-task workflow (v2)
 
 ## Overview
 
 4-phase chain mapping the user's CLAUDE.md Execute-phase discipline onto the harnessed
-runtime (ADR 0011 — SDK + ralph-loop integration):
+runtime (ADR 0011 — SDK + ralph-loop integration; v2 schema upgrade T2.4.W1.1):
 
-| phase | id | upstream | model (intel CD-2 § 第 4 条) | rationale |
+| phase | id | upstream | model (intel CD-2 § 第 4 条) | v2 wiring |
 | ----- | -- | -------- | ---------------------------- | --------- |
-| 1 | `01-clarify` | `superpowers brainstorming` | opus | 任务复杂度澄清 |
-| 2 | `02-code` | `karpathy` 心法 always-on | sonnet | small surgical changes |
-| 3 | `03-test` | `superpowers TDD` + mattpocock 招式 | sonnet | conditional TDD red-green-refactor |
-| 4 | `04-deliver` | `ralph-loop` | haiku | 迭代验收循环省 token (关键点) |
+| 1 | `01-clarify` | `superpowers brainstorming` | opus | `capability: superpowers-brainstorming` + `gate: judgments.subtask-gate.brainstorming.fires` |
+| 2 | `02-code` | `karpathy` 心法 always-on | sonnet | `on[]` route to tdd-gate / grill-with-docs / zoom-out (D-09 + D-13) |
+| 3 | `03-test` | `superpowers TDD` | sonnet | `capability: tdd` + `on[]` route to diagnose on test_fail (D-13 + D-09) |
+| 4 | `04-deliver` | `ralph-loop` | haiku | `capability: ralph-loop` + `args: {completion_promise: COMPLETE}` + `gate/parallelism: judgments.parallelism-gate.*` + `fallback.max_iterations_exceeded: emit_warning_and_halt` (R20.10) |
+
+v2 schema fields per `src/workflow/schema/workflow.ts` (T2.4.W0.1 16th surface — harnessed.workflow.v2):
+- `capability: '{{ capabilities.<name>.cmd }}'` template interpolation (D-10 capability abstraction)
+- `gate: judgments.<file>.<gate>.fires` 4-level ref (pre-resolved by T2.3.W0.4 judgmentResolver)
+- `on: [{if, invoke|action}]` conditional clause (D-09 mattpocock route by condition)
+- `args: {completion_promise, max_iterations}` ralph-loop SDK params (R20.10 verbatim COMPLETE gate)
+- `parallelism: judgments.parallelism-gate.<route>.fires` (D-11 subagent / Agent Teams / main session route)
+- `fallback.max_iterations_exceeded: {action, message, exit_code}` (R20.10 acceptance c "explicit NOT silent")
 
 Per-phase models load from `workflows/execute-task/phases.yaml`; engine.runRouting
 spawns each phase as a sub-agent via `@anthropic-ai/claude-agent-sdk` 0.3.142+
-(`AgentDefinition` 5-字段 unpack — ADR 0011 § 4).
+(`AgentDefinition` 5-字段 unpack — ADR 0011 § 4). ralph-loop SDK wrapper at 04-deliver
+reuses sister Phase 2.2 v0.2.0 ship: `src/routing/lib/ralphLoop.ts` (54L) + `sdkSpawn.ts` (91L)
++ 4-layer dual-signal `isComplete` (NOT 重写 — per RESEARCH § 3.1).
 
 ## CLI invocation (the only enforced entry — B-28)
 

package/workflows/execute-task/phases.yaml

@@ -1,27 +1,73 @@
-# execute-task workflow — per-phase model tier (ADR 0011 errata / intel CD-2 § 第 4 条).
-# Default table sourced from `.planning/intel/omc-comparison.md` § CD-2 (phase 2.2 W3 — T3.3).
+# workflows/execute-task/phases.yaml — Phase v2.0-2.4 W1.1 v2 schema upgrade
+# Sister v1 (28L, ADR 0011 errata Phase 2.2 W3 T3.3 ship) → v2 (~60L, T2.4.W1.1).
+#
+# v2 delta (per PLAN.md L327-330 verbatim + RESEARCH § 3.3 + D-09 + D-10 + D-13):
+#   - schema_version: harnessed.workflow.v2 (sister Phase 2.2 schemaVersion 7 surface)
+#   - 02-code:    on[] mattpocock route by condition (tdd-gate / grill-with-docs / zoom-out)
+#   - 03-test:    on[] diagnose if test_fail
+#   - 04-deliver: capability + args + gate + parallelism + on[] + fallback (R20.10 explicit halt)
+#
+# Per-phase model tier (ADR 0011 errata / intel CD-2 § 第 4 条) — unchanged from v1.
+# Default table sourced from `.planning/intel/omc-comparison.md` § CD-2 (phase 2.2 W3 T3.3).
 # Override via CLI: `--model-tier inherit` (B-10 escape hatch).
+#
+# capability template — '{{ capabilities.<name>.cmd }}' resolves at workflow runtime
+# (sister D-02 gstack_prefix interpolation pattern from workflows/plan-feature/workflow.yaml).
+# gate / parallelism — 4-level ref 'judgments.<file>.<gate>.fires' pre-resolves via
+# T2.3.W0.4 judgmentResolver BEFORE expr-eval (per src/workflow/schema/workflow.ts L18-19).
+
+schema_version: harnessed.workflow.v2
 workflow: execute-task
+description: 子任务执行 workflow — 4-phase 真接 ralph-loop SDK + tdd-gate conditional + mattpocock route by condition
 phases:
   - id: 01-clarify
     name: brainstorming
     upstream: superpowers brainstorming
-    model: opus            # intel 第 4 条: opus / sonnet — 任务复杂度澄清
-    skills: ['brainstorming']
-    max_iterations: 5
+    capability: '{{ capabilities.superpowers-brainstorming.cmd }}'
+    model: opus
+    max_iterations: '{{ defaults.ralph_max_iterations.execute-task.01-clarify }}'
+    gate: judgments.subtask-gate.brainstorming.fires
+
   - id: 02-code
-    name: code (karpathy 心法 always-on)
+    name: code (karpathy 心法 always-on + mattpocock conditional route)
     upstream: karpathy
     model: sonnet
-    max_iterations: 20
+    max_iterations: '{{ defaults.ralph_max_iterations.execute-task.02-code }}'
+    on:
+      - if: 'judgments.tdd-gate.tdd-strongly-suggested.fires'
+        invoke: '{{ capabilities.tdd.cmd }}'
+      - if: 'phase.spec_ambiguous == true'
+        invoke: '{{ capabilities.grill-with-docs.cmd }}'
+      - if: 'phase.unfamiliar_module == true'
+        invoke: '{{ capabilities.zoom-out.cmd }}'
+
   - id: 03-test
-    name: test (conditional TDD + mattpocock 招式)
+    name: test (conditional TDD + diagnose on fail)
     upstream: superpowers TDD
-    model: sonnet          # intel 第 4 条: sonnet / haiku — TDD red-green-refactor
-    skills: ['test-driven-development']
-    max_iterations: 15
+    capability: '{{ capabilities.tdd.cmd }}'
+    model: sonnet
+    max_iterations: '{{ defaults.ralph_max_iterations.execute-task.03-test }}'
+    on:
+      - if: 'test_fail == true'
+        invoke: '{{ capabilities.diagnose.cmd }}'
+
   - id: 04-deliver
-    name: deliver (ralph-loop COMPLETE)
+    name: deliver (ralph-loop COMPLETE gate + max-iter fallback)
     upstream: ralph-loop
-    model: haiku           # intel 第 4 条: 04-deliver=haiku (省 token — 迭代验收循环关键点)
-    max_iterations: 20
+    capability: '{{ capabilities.ralph-loop.cmd }}'
+    model: haiku
+    args:
+      completion_promise: COMPLETE
+      max_iterations: '{{ defaults.ralph_max_iterations.execute-task.04-deliver }}'
+    gate: judgments.parallelism-gate.fires
+    parallelism: judgments.parallelism-gate.ralph-loop-wrapper.fires
+    on:
+      - if: 'subtask.lines >= 20 and subtask.type != "single_command_query"'
+        invoke: '{{ capabilities.ralph-loop.cmd }}'
+      - if: 'subtask.lines < 20 or subtask.type == "single_command_query"'
+        action: skip
+    fallback:
+      max_iterations_exceeded:
+        action: emit_warning_and_halt
+        message: '⚠️ ralph-loop max-iterations ({{ args.max_iterations }}) exceeded for execute-task 04-deliver. Sub-task likely incomplete — see workflow engine catch handler for manual options.'
+        exit_code: 1