linkding-ingest

分类: 内容与多媒体 | 上传者: lmorchard | 下载: 0 | 版本: v1.0（最新）
获取最近的 Linkding 书签、阅读其内容并将见解记录到保管库
更新日志: Source: GitHub https://github.com/lmorchard/decafclaw
目录结构

当前层级: tree/main/contrib/skills/linkding-ingest/
📄 download-binary.sh 1.4 KB
📄 fetch.sh 1.3 KB
📄 SKILL.md 6.5 KB
SKILL.md

---
name: linkding-ingest
description: Fetch recent Linkding bookmarks, read their content, and record insights to the vault
schedule: "45 */4 * * *"
effort: default
required-skills:
  - tabstack
allowed-tools: shell($SKILL_DIR/fetch.sh), vault_read, vault_write, vault_search, vault_list, vault_backlinks, tabstack_extract_markdown, vault_journal_append, current_time, delegate_task
user-invocable: true
---

# Linkding Bookmark Ingestion

Fetch recent bookmarks from Linkding, then delegate each bookmark to a child agent for content extraction and vault integration.

## Output Folder — READ THIS FIRST

**All new bookmark-derived pages MUST be created under `agent/pages/bookmarks/`.** Never call `vault_write` with a page path that doesn't start with `agent/pages/bookmarks/`. Example valid paths:

- `agent/pages/bookmarks/Rust Async Runtimes`
- `agent/pages/bookmarks/tools/ripgrep`
- `agent/pages/bookmarks/programming/FFI patterns`

Use sub-organization by topic when it makes sense (e.g. `bookmarks/programming/`, `bookmarks/tools/`). Include `[[wiki-links]]` back to related pages elsewhere in the vault.

If `vault_search` finds a relevant page that's already under `agent/pages/` but outside `agent/pages/bookmarks/` (e.g. a page you or another skill put in `agent/pages/topics/foo`), update it in place — don't create a duplicate. If you find a relevant page *outside* `agent/pages/` (e.g. the user's own notes under a different vault root), do NOT modify it — create a new page under `agent/pages/bookmarks/` and optionally link to the user's page with a `[[wiki-link]]`.

## Configuration

Required environment variables (set in `.env` or `config.json` env section):

| Env Var | Description |
|---------|-------------|
| `LINKDING_URL` | Linkding instance URL (e.g. `https://links.example.com`) |
| `LINKDING_TOKEN` | API token (Linkding Settings > Integrations) |

## Process

### 1. Fetch the bookmark list

Run the fetch script:

```
$SKILL_DIR/fetch.sh
```

This outputs ALL recent bookmarks as markdown. Read the entire output to get the full list.

### 2. Delegate each bookmark

For EACH bookmark in the list, use `delegate_task` to spawn a child agent that will:
- Fetch the full article content
- Analyze it for key insights
- Update the wiki

The task description for each delegate should include:

```
Process this bookmark and update the wiki knowledge base:

URL: {bookmark_url}
Title: {bookmark_title}
Tags: {bookmark_tags}
Description: {bookmark_description}

You have these tools available:
- tabstack_extract_markdown(url) — fetch full article content
- vault_search(query) — search vault pages by name/content
- vault_read(page) — read a vault page (parameter is "page", not "path")
- vault_write(page, content) — create or overwrite a vault page (parameters are "page" and "content")
- vault_backlinks(page) — find pages linking to a page

PATH RULE: All NEW pages you create MUST have a `page` argument starting with `agent/pages/bookmarks/`. This is not a suggestion — any `vault_write(page=...)` that doesn't start with that prefix is wrong.

If `vault_search` returns an existing page already under `agent/pages/` (anywhere in that tree), update it in place rather than creating a duplicate. If it returns a page OUTSIDE `agent/pages/` (e.g. the user's own notes), do NOT modify it — create a new page under `agent/pages/bookmarks/` and link to the user's page with `[[wiki-link]]` instead.

FRONTMATTER RULE: Every page you write (new or updated) MUST begin with Obsidian-compatible YAML frontmatter containing a `tags` list. Seed the list from the Tags field above; add more tags if the content reveals clearer categories. When updating an existing page that already has frontmatter, merge the bookmark's tags into the existing `tags` list (preserve existing tags, deduplicate, keep other frontmatter fields intact). Shape:

```
---
tags:
  - topic-one
  - topic-two
---

# Page Title

...body...
```

Instructions:
1. Use tabstack_extract_markdown(url="{bookmark_url}") to fetch the full content. If it fails (paywall, dead link), work with just the title, tags, and description above.
2. Analyze the content for key facts, insights, technologies, people, projects, or concepts.
3. Use vault_search(query="relevant topic") to find existing vault pages.
4. If a relevant page exists: vault_read(page="...full existing path...") to get it, revise with new info, merge the bookmark's tags into the page's frontmatter `tags` list (dedupe), vault_write(page="...same existing path...", content="...") to save. Keep the existing path — don't move the page.
5. If no relevant page exists and the topic is substantial: vault_write(page="agent/pages/bookmarks/New Page", content="...") with frontmatter (including `tags:` seeded from the bookmark tags), [[wiki-links]], and a ## Sources section. The `page` argument MUST start with `agent/pages/bookmarks/`.
6. Include the original URL and {bookmark_date} in ## Sources.
7. Extract knowledge — "X uses Y approach for Z problem" is better than "bookmarked an article about X".
8. Prefer adding to existing vault pages over creating new ones — but if you must create, use the `agent/pages/bookmarks/` prefix.
9. Do NOT use tool_search — it is not available. Use only the tools listed above.
```

You can delegate multiple bookmarks concurrently — `delegate_task` runs them in parallel.

Skip obviously low-signal bookmarks (duplicates, ephemeral content) — don't waste a delegate on them.

### 3. Finish

After all delegates complete, end with a short narrative summary of what was processed and what wiki pages were updated or created. If there was nothing interesting to ingest this cycle, begin your summary with `HEARTBEAT_OK` on its own line followed by a brief quiet-cycle note — the leading marker lets the scheduler log a tidy line, and the narrative keeps the archive readable for the newsletter.

## Rules

- **New pages go under `agent/pages/bookmarks/`** — never create a new page at the vault root. This applies to the parent turn and to every delegate you spawn.
- **Frontmatter with `tags:` is required on every written page** — seed from the bookmark's tags; merge when updating an existing page. Tags boost semantic search and future tag-based retrieval.
- **Delegate each bookmark** — don't try to fetch and process articles yourself, your context will fill up
- **Group related content** — include guidance in the delegate task to add to existing pages when possible
- Convert relative dates to absolute dates
- Include original URLs so sources can be revisited
登录后下载/点赞/收藏 ❤ 10 | ★ 0
linkding-ingest

目录结构

SKILL.md

举报内容

提示
