Doc Web Assistant

---

name: doc-web-assistant

description: "Use native web_fetch content to build a local documentation knowledge base, query it, extract command plans, and prepare Doc Executor workflows."

metadata:

---

# Doc Web Assistant

Purpose

This skill lets OpenClaw treat a documentation website as a local structured knowledge base.

Read When

Runtime Requirements

Allowed Tools

It supports three stages:

1. Use OpenClaw native `web_fetch` to obtain page content.

2. Import fetched content into local JSON files.

3. Query the local JSON store by natural language.

4. Build an execution plan from matched documentation commands.

The acquisition path is:

1. Use native `web_fetch` first.

2. Save or pass the fetched content into this skill via `import`.

This skill is designed for documentation assistant workflows such as:

Skill files

Installation

Install the required Python packages:

pip install -r requirements.txt

If OpenClaw runs inside the skill directory, this is enough.

If you need an explicit command with absolute path, use:

python doc_web_assistant.py --help

Core commands

1. Import fetched content into JSON

Preferred path for OpenClaw.

First, use native `web_fetch` to get the document content from the target URL.

Then save that content to a local file such as `./fetched_doc.md` or `./fetched_doc.html`.

Finally import it with:

python doc_web_assistant.py import --input "./fetched_doc.md" --source-url "https://example.com/docs/install" --out "./doc_store"

You may also import HTML directly:

python doc_web_assistant.py import --input "./fetched_doc.html" --source-url "https://example.com/docs/install" --out "./doc_store" --format html

Output files will be created under `./doc_store`:

2. Query imported documentation

python doc_web_assistant.py query --db "./doc_store" --query "如何安装 ARMv9 优化版 llama.cpp" --top-k 5

This returns the most relevant sections and code blocks from the local store.

3. Build a doc-grounded command plan

python doc_web_assistant.py plan --db "./doc_store" --query "根据文档给我安装 KleidiAI 优化版 llama.cpp 的命令" --top-k 5

This returns:

Recommended OpenClaw workflow

When the user asks a documentation-driven task, follow this sequence.

Workflow A: First-time doc import with native fetch

Use when the target documentation is not yet available locally.

1. Use OpenClaw native `web_fetch` to fetch `<DOC_URL>`.

2. Save the fetched content to a temporary file.

3. Import that file into the local doc store.

python doc_web_assistant.py import --input "./fetched_doc.md" --source-url "<DOC_URL>" --out "./doc_store"

Then inspect results with:

python doc_web_assistant.py query --db "./doc_store" --query "<USER_TASK>" --top-k 5

Workflow B: Answer from docs

Use when the user wants explanation only.

python doc_web_assistant.py query --db "./doc_store" --query "<USER_QUESTION>" --top-k 5

Answer from the matched chunks and cite the source URL when possible.

Workflow C: Plan terminal operations from docs

Use when the user wants executable steps.

python doc_web_assistant.py plan --db "./doc_store" --query "<USER_TASK>" --top-k 5

Then convert returned commands into an ordered plan.

Workflow D: Execute after planning

Use only after a plan is built.

Operational policy for OpenClaw

When using this skill, OpenClaw should behave as follows.

If the user says

Then OpenClaw should

1. Determine whether the task needs import, query, or plan.

2. If the task references a URL and no local store exists, use native `web_fetch` first.

3. Import the fetched content with `python doc_web_assistant.py import ...`.

4. Use `query` for explanation tasks.

5. Use `plan` for terminal-execution tasks.

6. Present or execute commands according to risk level.

7. Prefer commands extracted from documentation instead of inventing commands.

8. If native `web_fetch` returns incomplete or noisy content, say so explicitly before planning execution.

JSON schema overview

`manifest.json`

Contains:

`pages/page_xxxx.json`

Contains per-page structured content:

`chunks.json`

Contains flattened retrieval chunks for:

Example: build a local doc assistant

Native fetch + import

1. Use native `web_fetch` on:

2. Save the fetched content to `./radxa_llama_doc.md`

3. Import it:

python doc_web_assistant.py import --input "./radxa_llama_doc.md" --source-url "https://docs.radxa.com/en/orion-o6/app-development/llama.cpp" --out "./radxa_llama_docs"

Query

python doc_web_assistant.py query --db "./radxa_llama_docs" --query "KleidiAI 编译参数" --top-k 3

Plan

python doc_web_assistant.py plan --db "./radxa_llama_docs" --query "根据文档生成安装和编译命令" --top-k 5

Safety rules

Notes

Troubleshooting

Reporting issues

If the skill logic is unclear, update `SKILL.md` and keep the command examples aligned with the Python script.