メニュー
Load when the user wants to ingest a URL, document, or text snippet into the knowledge base for future retrieval. Trigger phrases: "save this", "ingest", "add to KB".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Load when the user asks to queue, enqueue, or add a coding task to the fleet, dispatch a job to a repo, or ask the fleet worker to fix something. Trigger phrases: "add to fleet", "queue a job for", "send to fleet", "dispatch to".
Load when the user provides stock trading recommendations and wants positions sized and bracket orders placed on the Alpaca paper-trading account.
Load when the user wants to navigate a website, click through a flow, fill a form, extract data from a page, or record an interaction as a reusable playbook.
Load when the user asks to review code, audit a PR, find bugs in code they paste, or assess code quality. Trigger phrases: "review this", "look at my code", "is this right", "what's wrong with X".
Load when the user is responding to a customer inquiry, complaint, or support ticket and needs help drafting a reply with the right tone.
Load when the user wants analysis on a dataset — trends, distributions, summary statistics, or insights from a CSV/JSON/Excel file.
| name | Knowledge Ingestion |
| slug | knowledge-ingestion |
| description | Load when the user wants to ingest a URL, document, or text snippet into the knowledge base for future retrieval. Trigger phrases: "save this", "ingest", "add to KB". |
| icon | Upload |
| color | #059669 |
| version | 1.1.0 |
| category | research |
| tools | ["ingest_url_to_kb","scrape_webpage","search_knowledge_base"] |
| config_schema | {"type":"object","properties":{"target_kb_id":{"type":"string","default":""},"chunk_strategy":{"type":"string","enum":["page","section","paragraph"],"default":"section"}}} |
Target KB: {target_kb_id}. Chunk strategy: {chunk_strategy}.
Before ingesting, search the KB for the same source — duplicate ingestion creates noisy retrieval.
For URLs: ingest_url_to_kb directly.
For pages requiring navigation (login, JS): scrape_webpage first, then ingest the extracted text.
After ingest, verify by searching for a distinctive phrase from the document. If it doesn't surface, the ingestion silently failed.
Flag stale-prone sources (pricing pages, news, changelogs) in the title so they can be re-ingested later.
