メニュー
Use when refining agent context: compressing long notes, extracting durable facts, separating signal from noise, preparing task briefs, or identifying what context should be carried into a new session.
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
| name | context-refinery |
| description | Use when refining agent context: compressing long notes, extracting durable facts, separating signal from noise, preparing task briefs, or identifying what context should be carried into a new session. |
| version | 1.0.0 |
| triggers | ["refine this context","compact these notes","prepare a task brief","extract durable facts"] |
| negative_triggers | ["generate UI art","deploy a service"] |
| allowed_tools | [] |
| required_tools | [] |
| trust_level | local |
| safety_class | read_only |
| examples | ["Turn a long thread, note set, or context pack into a concise brief with facts, open questions, and next actions."] |
| tests | ["sb skills validate --path skills --strict"] |
| performance | {"notes":"Best for context compaction, salience filtering, and handoff briefs."} |
Transform noisy working context into a compact, useful brief without losing important constraints or provenance.
Process a meeting transcript to extract action items (with owners and due dates), risks, follow-up questions, and a concise summary. Optionally record decisions and ingest the output into the vault.
Search and synthesize knowledge from SecondBrain vault. Use when the user asks about their projects, past decisions, meetings, people, goals, or anything that might be in their personal knowledge base. Triggers on questions like "what do I know about X", "find notes on Y", "what did we decide about Z", "remind me about", "what's the status of".
Answer data and analytics questions using SecondBrain DataAgent — SQL execution, workflow runs, and LLM synthesis over the knowledge database. Use for questions about metrics, counts, trends, comparisons, or anything that sounds like "how many", "show me the data", "analyse", "what are the trends in".
Extract decisions from a markdown file and help record them in the SecondBrain decision catalog with proper namespace keys.
Answer a research question using closed-corpus grounded reasoning with evidence chains and citations. Use for deep research questions that need traceable evidence, cross-document synthesis, or when the user asks to "research", "investigate", or "dig into" something thoroughly.
Ingest a file or directory into SecondBrain vault — parse, chunk, embed, and index it for future retrieval. Use when the user wants to add content to their knowledge base.
