hermes-task-framework 1.0.0__py3-none-any.whl

task-framework/skills/task-framework/references/research-workflow.md

@@ -0,0 +1,220 @@
+# Parallel Research Workflow
+
+Use delegate_task with parallel subagents for comprehensive research tasks. Each subagent covers one well-scoped direction and produces structured markdown files.
+
+## When to Use
+
+- Market research / competitive analysis
+- Policy and regulatory research
+- Technology landscape scanning
+- Multi-angle topic investigations
+- Any research that can be naturally split into independent streams
+
+## Phased Research Lifecycle
+
+Complex research projects benefit from a phased approach where each phase produces output that feeds the next:
+
+| Phase | Focus | Delegation pattern | Output |
+|-------|-------|-------------------|--------|
+| 1 — Broad scoping | Initial search, landscape overview | Inline (manual search) | README.md + 2-3 docs/ files |
+| 2 — Deep dive | Competitor/product deep research | 1 subagent (multiple competitors serial) | 5-6 competitor-*.md files |
+| 3 — Comprehensive | Policy + patents + keywords + ecosystem | 3 parallel subagents | 5+ docs covering all dimensions |
+| 4 — Technical feasibility | Deep technical analysis | 1 deep subagent | 1 large comprehensive doc |
+| 5 — Synthesis | Cross-document compilation | 1 synthesis subagent reading all prior docs | Final product plan doc |
+
+### Phase transition flow
+
+After each phase completes:
+
+1. Read the subagent summary (not just the file — check the summary section)
+2. Verify files were written: `ls -lh docs/`
+3. Mark the phase `[x]` in TASK.md
+4. Update README.md with a summary table of new docs
+5. Present cross-document insights to the user
+6. If the user provides additional guidance (new keywords, new angles), capture it in TASK.md's Notes section, then adjust the next phase's scope
+7. Ask "continue to next phase?" before proceeding
+
+## Delegation Patterns by Phase Type
+
+### Broad scoping (Phase 1)
+**Strategy:** Inline. Do the searches yourself, compile into docs manually.
+
+### Multi-competitor deep dive (Phase 2)
+**Strategy:** One subagent, many targets. Give it a list of 5-8 competitors and let it work through them serially within one delegate_task call. Each competitor gets its own file.
+
+```python
+delegate_task(
+    goal="Research 6 competitors in this market",
+    context="""...
+    # Save each as docs/competitor-<name>.md
+    # Each file must end with a 启示 (Implications) section
+    """,
+    toolsets=["terminal", "file", "browser"]
+)
+```
+
+### Multi-dimensional research (Phase 3)
+**Strategy:** Parallel subagents, each covering a different dimension. Split by natural fault lines (policy / technology / products / ecosystem), not by source.
+
+```python
+delegate_task(
+    tasks=[
+        {"goal": "Research policy & industry reports", ...},
+        {"goal": "Research patents & expanded keywords", ...},
+        {"goal": "Research companion products & startups", ...},
+    ]
+)
+```
+
+Each subagent produces 1-2 docs. The content should be non-overlapping.
+
+### Deep technical analysis (Phase 4)
+**Strategy:** One deep subagent with a comprehensive spec. Give it 7-8 required sections as a template, lots of context from prior phases.
+
+### Synthesis / product planning (Phase 5)
+**Strategy:** One synthesis subagent with instructions to READ all prior docs, then produce the final document. Grant `file` toolset so it can read existing material.
+
+```python
+delegate_task(
+    goal="Synthesize all research into product plan",
+    context=f"""Read all docs from docs/ directory, then produce product-plan-draft.md
+    with 8 chapters...""",
+    toolsets=["terminal", "file"]
+)
+```
+
+## Split the research domain
+
+Identify natural fault lines and split into parallel tasks:
+
+| Good split | Bad split |
+|-----------|-----------|
+| Competitor A / Competitor B / Competitor C | Feature 1 of A / Feature 2 of A / History of A |
+| Policy / Patents / Products | Same source queried 3 ways |
+| Domestic / International / Technology | Overlapping company lists |
+
+Each task must be **independent** — no shared intermediate state.
+
+## Write task specs with explicit output directives
+
+Each subagent's prompt must include:
+
+```
+1. Research scope (what to search)
+2. Output file path (absolute, where to save)
+3. File format (sections, required headers)
+4. Naming convention (prefix conventions like `competitor-`, `paper-`, `policy-`)
+```
+
+Example task spec:
+
+```
+"context": "输出文件：~/studio/project/tasks/*.research/docs/competitor-qinjia.md
+格式规范：
+# 亲家科技
+## 产品体系
+(功能模块列表)
+## 目标客户
+## 技术栈
+## 案例
+## 启示（对本项目的参考价值）"
+```
+
+## Run parallel subagents
+
+```python
+delegate_task(
+    tasks=[
+        {"goal": "...", "context": "...", "toolsets": ["terminal", "file", "browser"]},
+        {"goal": "...", "context": "...", "toolsets": ["terminal", "file", "browser"]},
+        {"goal": "...", "context": "...", "toolsets": ["terminal", "file", "browser"]},
+    ]
+)
+```
+
+Max concurrent tasks for this user: 3. Time budget: ~5-10 min each.
+
+## Compile results
+
+After all subagents return:
+
+1. Read the summary sections of each result
+2. Verify files were actually written (`ls -lh docs/`)
+3. Update `TASK.md`: mark the phase as `[x]`
+4. Update `README.md`: add a summary table of collected docs
+5. Present cross-document insights to the user
+
+## Output File Naming Convention
+
+| Prefix | Use case |
+|--------|----------|
+| `competitor-<company>.md` | One company/competitor per file |
+| `paper-<short-title>.md` | Academic paper review |
+| `policy-<topic>-<year>.md` | Policy/regulatory research |
+| `patent-<direction>.md` | Patent landscape |
+| `extended-keywords-<topic>.md` | Keyword-expanded product discovery |
+| `companion-products.md` | Companion/care products |
+| `startup-bp-landscape.md` | Startup business model analysis |
+| `industry-report-<topic>.md` | Market data and forecasts |
+| `architecture-proposal.md` | Technical architecture design |
+| `tech-feasibility-analysis.md` | Technology stack evaluation |
+| `product-plan-draft.md` | Final product plan |
+
+Each file should include a "启示" (Implications) section at the end showing relevance to the project.
+
+## Research Methodology Tips
+
+### Keyword expansion
+
+Start with a core term and snowball outward. Example:
+
+```
+康养
+  ├── 银发经济
+  ├── 智慧助老
+  ├── 居家适老化改造
+  ├── 慢病管理
+  ├── 医养结合
+  ├── 社区养老
+  ├── 老年人陪护
+  └── AI陪聊
+```
+
+Each expanded keyword surfaces a different set of products, companies, and policy frameworks. Review the user's guidance mid-project — they may suggest additional keywords that were not in the original scope. Capture these in TASK.md's Notes section and incorporate into the next phase.
+
+### Multi-source strategy
+
+| Source | Type | What to look for |
+|--------|------|-----------------|
+| arXiv | Academic | Foundation research, emerging techniques |
+| 中国专利数据库 | Patent | Productized technology, applicant landscape |
+| 36氪/IT桔子 | Startup | Funding rounds, business models, pitch decks |
+| 政府网站 (gov.cn) | Policy | Directives, pilot programs, subsidy criteria |
+| Company websites | Product | Feature lists, pricing, case studies, team |
+| 工信部目录 | Directory | Approved products, vendor lists |
+
+### Output quality tracking
+
+After each phase, log the cumulative document count and line count so the user can see progress:
+
+```bash
+find docs/ -name '*.md' -exec wc -l {} + | sort -rn
+du -sh .
+```
+
+### User guidance integration mid-project
+
+When the user provides additional research directions mid-project (common in exploratory research):
+
+1. Add the new directions to TASK.md as a bullet list under "## Notes" or a dedicated "## 新增调研方向" section
+2. Update the next phase's scope to incorporate them
+3. Don't restart — fold into the next delegation call
+
+### Pitfalls
+
+- **Chinese search engines block bots** — Bing.cn with `&cc=cn` works sometimes, but expect 50% failure. Use gov.cn and 36氪 as fallback sources.
+- **Company websites return 403** — some older Chinese sites block non-China IPs. Use cached pages or third-party descriptions.
+- **Pricing is rarely public** — Chinese B-end SaaS products rarely list prices. Infer from case study size and funding data.
+- **Subagent summaries can be wrong** — always verify file writes with `ls -lh` after delegation returns.
+- **Subagent output limits** — very large research docs (>800 lines) may approach context limits. Split into 2 documents if a single subagent produces huge output.
+- **Parallel task independence** — subagents cannot read each other's output mid-run. Each must be fully self-contained with all context provided in the prompt.

task-framework/skills/task-framework/references/task-format-validation.md

@@ -0,0 +1,25 @@
+# TASK.md Format Validation Checklist
+
+Systematic checklist for validating a TASK.md against the `project-tasks` skill spec.
+
+## Section Checklist
+
+- [ ] **`# Task: <Name>`** — Title matches the task directory name (after the timestamp prefix)
+- [ ] **`## Status`** — Header present. Next line is `active | paused | completed | cancelled — <reason>`. Parsable by `grep -m1 '^## Status'`
+- [ ] **`## Goal`** — Single one-liner describing what the task achieves
+- [ ] **`## Pre-checks`** — Section name is EXACTLY `## Pre-checks` (not `Pre-check results`, `Pre-checks (executed X)`, etc.). The executor checks for this exact string. If the section exists, checkboxes follow directly under it (not under a subheading like `### 待决策事项`)
+- [ ] **`## 环境要求`** — Table format with at minimum columns: `项目`, `最低要求`, `说明`. Present when the task has environment dependencies
+- [ ] **`## Checklist`** — Phase headings (`### Phase N — ...`) with `[ ]` / `[x]` items. Optional `BREAK:` lines between phases
+- [ ] **`## Notes`** — Free-form section at the bottom. Contains decisions, reference links, or research summaries that don't belong in Pre-checks
+
+## Common Deviations (check for these)
+
+| Deviation | Fix |
+|-----------|-----|
+| `## Pre-check results` or `## Pre-checks results` | Rename to `## Pre-checks` |
+| Pre-check checkbox items under a subheading (`### 待决策事项`) | Move checkboxes directly under `## Pre-checks` |
+| Research notes/results inline in Pre-checks section | Move detailed findings to `## Notes`; keep Pre-checks as compact checkboxes only |
+| Stray table-syntax artifacts (`|## Pre-checks`) | Remove leading `|` and trailing `|` from section headers |
+| Status line missing the dash format | Rewrite as `active — <reason>` (space, em-dash, space) |
+| `BREAK:` line inside a code block | BREAK must be unformatted markdown text, not inside backticks |
+| No `## 环境要求` when task needs specific tools/API keys/env vars | Add the table with accurate minimum requirements |

task-framework/skills/task-framework/references/task-hash-naming.md

@@ -0,0 +1,40 @@
+# Task Hash Naming Convention
+
+## Why the hash must be in the directory name
+
+`resolve_ref()` in `scripts/task_ref.py` locates tasks by globbing `*{hash_id}*` against the `tasks/` directory:
+
+```python
+matches = glob.glob(os.path.join(TASKS_ROOT, f'*{hash_id}*'))
+```
+
+This matches **directory names**, not file contents. If the hash is only in `.hermes-task.json` and not in the directory name, the glob returns nothing and `ref:` lookups fail with `FileNotFoundError`.
+
+## The convention
+
+```
+tasks/YYYYMMDD-HHMMSS.<name>-<hash6>/
+```
+
+The hash suffix at the end of the directory name ensures:
+- `resolve_ref('ref:<hash6>/...')` can find the task by matching `*<hash6>*`
+- The timestamp + name part remains human-readable
+- The hash is globally unique, so no collisions even with identical task names
+
+Example:
+```
+tasks/20260605-233355.health-sales-demo-5d5a1a/
+├── .hermes-task.json     ← hash: "5d5a1a", name: "health-sales-demo-5d5a1a"
+├── TASK.md
+└── ...
+```
+
+## What NOT to do
+
+- ❌ Do NOT omit the hash from the dir name — `ref:` resolution breaks
+- ❌ Do NOT use sequential version suffixes (`-v1`, `-v2`) — they're not hash-based and convey no identity
+- ❌ Do NOT `mv` task directories manually — use task-framework tools. Manual rename breaks the `ref:` system across all referring tasks until the directory name is updated to match
+
+## Consistency with phase directories
+
+Phase output directories use the same `<short-name>-<hash6>/` pattern (e.g. `tts-a3f8c2/`, `compositing-c9f3a2/`). The task directory is just the top-level version of the same convention: `<ts>.<name>-<hash6>/`.

task-framework/skills/task-framework/references/task-lifecycle-example.md

@@ -0,0 +1,122 @@
+# Task Lifecycle Walkthrough
+
+This example traces a paper-reproduction task from creation through completion.
+
+## 1. Create
+
+For code/tool tasks, create logs/ only:
+
+```
+ts=$(date '+%Y%m%d-%H%M%S')
+mkdir -p "tasks/${ts}.paper-reproduction/logs/"
+```
+
+For research/documentation tasks, also create docs/:
+
+```
+ts=$(date '+%Y%m%d-%H%M%S')
+mkdir -p "tasks/${ts}.kangyang-center-research/logs/"
+mkdir -p "tasks/${ts}.kangyang-center-research/docs/"
+```
+
+Result: `tasks/20260601-164001.paper-reproduction/` (code) or
+         `tasks/20260602-041236.kangyang-center-research/` (research)
+
+## 2. Status — active
+
+TASK.md starts with:
+
+```markdown
+## Status
+
+active — running pre-checks
+```
+
+## 3. Pre-checks (optional)
+
+Present when uncertainty exists:
+
+```markdown
+## Pre-checks
+
+- [ ] Can DeepSeek API replace GPT-4?
+- [ ] Can uv replace conda?
+```
+
+Agent:
+1. Reads Pre-checks section
+2. For each `[ ]`, researches and writes `logs/precheck.YYYYMMDD-HHMMSS.md`
+3. Marks `[ ]` → `[x] conclusion`
+4. When all done, presents summary and pauses for user confirmation
+
+After confirmation, `## Pre-checks` section is typically kept with `[x]` items (history) or deleted entirely.
+
+## Checklist with BREAK
+
+```markdown
+## Checklist
+
+- [x] Phase 1 — Environment setup          ← already done
+- [ ] Phase 2 — Run experiments
+- [ ] BREAK: Check initial results before proceeding
+- [ ] Phase 3 — Verify against paper
+```
+
+Agent iterates:
+- `[x]` → skip
+- `[ ] Phase 1` → skip (already checked)
+- `[ ] Phase 2` → execute, then continue
+- `[ ] BREAK: ...` → stop, output summary + message, wait for user
+
+## 5. User confirms BREAK
+
+Agent rewrites TASK.md:
+
+```markdown
+- [x] DONE: Check initial results before proceeding
+- [ ] Phase 3 — Verify against paper
+```
+
+Execution resumes on next `task_run`.
+
+## 6. Completion
+
+When all `[ ]` are `[x]`, the agent sets:
+
+```markdown
+## Status
+
+completed — all phases done, ASR matches paper
+```
+
+## Directory structure at end
+
+For a **code/tool task** (paper reproduction):
+
+```
+tasks/
+└── 20260601-164001.paper-reproduction/
+    ├── README.md
+    ├── TASK.md
+    └── logs/
+        ├── precheck.20260601-164001.md
+        ├── output.20260601-164500.log
+        ├── error.20260601-164500.log
+        ├── output.20260601-171200.log
+        └── error.20260601-171200.log
+```
+
+For a **research/documentation task** (康养中心调研):
+
+```
+tasks/
+└── 20260602-041236.kangyang-center-research/
+    ├── README.md              ← overview + key findings summary
+    ├── TASK.md                ← checklist (Phase 1-5)
+    ├── docs/                  ← organized markdown research materials
+    │   ├── paper-summaries.md
+    │   ├── product-categories.md
+    │   └── architecture-proposal.md
+    └── logs/
+        └── summary.20260602-120000.log
+```

task-framework/skills/task-framework/references/task-overview-discovery.md

@@ -0,0 +1,31 @@
+# Task Overview Discovery
+
+When asked to list or update a summary of all tasks, use this approach rather than relying on a potentially stale root README.md.
+
+## Approach
+
+1. Scan the `tasks/` directory for subdirectories matching `YYYYMMDD-HHMMSS.*` pattern
+2. For each, read the `README.md` to extract:
+   - **Title** (`# Title` — first H1)
+   - **Description** (first line after H1)
+   - **Status** (`## Status` section or inline mention)
+3. Present a compact table to the user
+
+## Fallback When README.md is Empty
+
+If a task directory has no README.md, check for `TASK.md` and extract the first H1 and status section.
+
+## Index File Maintenance
+
+The root `tasks/README.md` is the canonical index. After any read/update, regenerate it with the `update-task-index` script from the skill's SKILL.md.
+
+## Output Format
+
+Prefer a compact pipe-delimited table for terminal readability:
+
+```
+| Directory | Title | Status | Description |
+|-----------+-------+--------+-------------|
+```
+
+Avoid bullet lists or verbose paragraphs for listing purposes.

task-framework/skills/task-framework/references/task-types/analysis.md

@@ -0,0 +1,60 @@
+# analysis 任务生命周期
+
+## 创建
+
+1. 创建 `tasks/<ts>.<name>-<hash6>/` 目录，含 `input/` `output/` `scripts/`
+2. 从 inbox 复制源文件（PDF、DOCX、图片等）到 `input/`
+3. 写入 `TASK.md`，Data Flow 中将源文件路径写为 `input/<filename>`
+4. 写入 `.hermes-task.json`
+
+## 执行
+
+1. 从 `input/` 读取源文件
+2. 生成分析文档写入 `output/docs/`
+3. **如果 TASK.md 有 `## Repo` 字段：** 设计文档/分析报告放到 Repo 路径下（如 `{Repo}/docs/`），而不是 `output/docs/`
+4. 执行日志仍写入 `output/logs/`（日志始终在 task 目录）
+5. 最终交付物在上一步确定的路径中（Repo docs/ 或 output/docs/）
+
+示例目录结构：
+
+```
+tasks/<ts>.<name>-<hash6>/
+├── input/
+│   └── source.pdf
+├── output/
+│   ├── docs/
+│   │   ├── 01-pdf-analysis.md
+│   │   ├── 02-prd-overview.md
+│   │   └── 03-prd-modules.md
+│   └── logs/
+│       └── output.20260610-152823.log
+├── TASK.md
+├── README.md
+├── TASK_MEMORY.md
+└── .hermes-task.json
+```
+
+## 修改
+
+- 修改 TASK.md checklist 正常进行
+- 不需要重跑已完成的 phase，除非修改了对应的输入
+
+## 清理
+
+`task_reset --hard` 执行：
+
+```bash
+rm -rf output/            # 删除所有生成文档和日志
+# 重置 TASK.md checkboxes
+# 重置状态为 active
+```
+
+如果 task 有自定义清理需求（如保留特定目录），创建 `scripts/clean.sh` 覆盖默认行为。
+
+`input/` 不受影响。
+
+## 完成
+
+1. 确认所有交付物在 `output/docs/` 中
+2. 更新 TASK.md 状态为 `completed`
+3. 运行 `python3 ~/.hermes/skills/software-development/task-framework/scripts/update-index.py` 更新索引

task-framework/skills/task-framework/references/task-types/external-audit.md

@@ -0,0 +1,56 @@
+# external-audit 任务生命周期
+
+## 创建
+
+1. 创建 `tasks/<ts>.<name>-<hash6>/` 目录，含 `output/` `scripts/`（不需要 `input/`）
+2. 在 TASK.md 的 Data Flow 中列出**外部源文件路径**（绝对路径，如 `~/project/docs/*.md`）
+3. 写入 `TASK.md`，注明"不可修改外部文件"
+4. 写入 `.hermes-task.json`
+
+## 执行
+
+1. 通过绝对路径读取外部源文件（`read_file`, `search_files` 工具）
+2. 将分析产出写入 `output/docs/`
+3. **如果 TASK.md 有 `## Repo` 字段：** 分析报告/审计文档放到 Repo 路径下（如 `{Repo}/docs/`），而不是 `output/docs/`
+4. 执行日志仍写入 `output/logs/`（日志始终在 task 目录）
+5. 最终交付物在上一步确定的路径中（Repo docs/ 或 output/docs/）
+
+**🔴 核心纪律：不得以任何方式修改 input 中引用的外部路径下的文件。** 不允许 `write_file`、`patch`、`terminal(mv/cp/rm)` 等操作触及外部路径。
+
+示例目录结构：
+
+```
+tasks/<ts>.<name>-<hash6>/
+├── output/
+│   ├── docs/
+│   │   ├── 01-platform-audit.md
+│   │   └── 02-recommendations.md
+│   └── logs/
+│       └── output.20260610-152823.log
+├── TASK.md
+├── README.md
+└── .hermes-task.json
+```
+
+## 修改
+
+- 修改 TASK.md checklist 正常进行
+- 如果需要重新分析，重新读取外部文件即可
+
+## 清理
+
+`task_reset --hard` 执行：
+
+```bash
+rm -rf output/            # 删除所有分析文档和日志
+# 重置 TASK.md checkboxes
+# 重置状态为 active
+```
+
+外部文件不受影响（本来就不在任务目录内）。
+
+## 完成
+
+1. 确认所有交付物在 `output/docs/` 中
+2. 更新 TASK.md 状态为 `completed`
+3. 运行 `python3 ~/.hermes/skills/software-development/task-framework/scripts/update-index.py` 更新索引

task-framework/skills/task-framework/references/task-types/video-production-pipeline.md

@@ -0,0 +1,77 @@
+# video-pipeline 任务生命周期
+
+## 创建
+
+1. 通过 `task_create --skill browser-screen-record-task` 创建
+2. 创建 `tasks/<ts>.<name>-<hash6>/` 目录，含 `input/` `output/` `scripts/`
+3. 用户填写的 `REQUIREMENTS.md` 和 `images/`（封面图片）放在 `input/`
+4. `browser-screen-record-task` 的 `scripts/create_task.py` 生成初始 `TASK.md`
+5. 写入 `.hermes-task.json`
+
+## 执行
+
+完整的 5 阶段流水线，使用 `pipeline.py` 驱动：
+
+| Phase | 生成到 | 说明 |
+|-------|--------|------|
+| 0 | 自身 | 解析 REQUIREMENTS.md → 翻译（如需要） |
+| 1a | `output/tts-<hash6>/` | TTS 音频 |
+| 1b | `output/image-slideshow-<hash6>/` | 封面视频 |
+| 1c | `output/subtitle-gen-<hash6>/` | 字幕 |
+| 2 | `output/` | 计算时间线 → RECORDING.md + COMPOSITING.md |
+| 3 | `output/browser-video-recording-<hash6>/` | 浏览器录屏 |
+| 4 | `output/compositing-<hash6>/` | 合成最终视频 |
+
+示例目录结构（执行后）：
+
+```
+tasks/<ts>.<name>-<hash6>/
+├── input/
+│   ├── REQUIREMENTS.md
+│   └── images/
+│       └── cover.png
+├── output/
+│   ├── tts-a3f8c2/
+│   │   ├── audio_000001.mp3
+│   │   └── audio_manifest.json
+│   ├── image-slideshow-a3f8c2/
+│   │   └── slideshow.mp4
+│   ├── subtitle-gen-a3f8c2/
+│   │   └── subtitle_0001.mp4
+│   ├── browser-video-recording-a3f8c2/
+│   │   └── video.mp4
+│   ├── compositing-a3f8c2/
+│   │   └── output.mp4
+│   ├── RECORDING.md
+│   └── COMPOSITING.md
+├── TASK.md
+├── README.md
+├── TASK_MEMORY.md
+└── .hermes-task.json
+```
+
+> **迁移说明：** 现有 pipeline 任务（如 `health-sales-demo-5d5a1a`）可能还把 REQUIREMENTS.md 和 images/ 放在根目录。后续对此任务操作时，应迁移到 `input/` + `output/` 模型。
+
+## 修改
+
+- **禁止修改 REQUIREMENTS.md** — 需要改时告知用户，让用户自己改
+- 修改后需要重新执行 pipeline
+
+## 清理
+
+```bash
+# 方式 A（推荐）：通过 pipeline.py
+cd tasks/<ts>.<name>-<hash6>/
+python3 ~/.hermes/skills/media/video-production-pipeline/scripts/pipeline.py --clean
+
+# 方式 B：通用清理（input/ 安全）
+rm -rf output/
+```
+
+`input/` 中的 REQUIREMENTS.md 和 images/ 不受影响。
+
+## 完成
+
+1. 确认 `output/compositing-<hash6>/output.mp4` 存在且可播放（ffprobe 验证）
+2. 更新 TASK.md 状态为 `completed`
+3. 运行 `python3 ~/.hermes/skills/software-development/task-framework/scripts/update-index.py` 更新索引