메뉴
Hybrid retrieval (vector + FTS) over L2 structured summaries. Supports filter by domain, importance, thinking-stage, source, and extract mode. Use when user wants structured knowledge, not raw messages.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Evaluate a decision against the user's stated principles + related past thinking. Loads principles from YAML, matches by keyword, surfaces semantically similar prior decisions.
Raw accounting of what's in the brain. Counts messages, embeddings, summaries, domains. Use when user asks "how much is in my brain" or "what's the size of the corpus."
Surface patterns in how the user thinks — cognitive pattern, problem-solving approach, emotional tone, content category. Optional domain scope. Use when user asks "when do I think best" or "what patterns show up in breakthroughs."
Full re-entry brief for a domain — recent summaries + accumulated open questions, decisions, insights, quotes. Longer and deeper than tunnel-state. Use after days/weeks away from a project.
Browse conversations by calendar date. Use when user asks "what was I working on last Thursday" or specifies a date.
Surface abandoned domains with unresolved breakthrough/significant open questions. Ranked by breakthrough count → question count → conversation count. Use when user feels important threads have been dropped.
| name | search-summaries |
| description | Hybrid retrieval (vector + FTS) over L2 structured summaries. Supports filter by domain, importance, thinking-stage, source, and extract mode. Use when user wants structured knowledge, not raw messages. |
| layer | L1 |
| reads | ["L2.summaries"] |
| writes | [] |
| citations | output-is-citation |
| determinism | pure-function |
| allowed-tools | mcp__my-brain__search_summaries |
L1 retrieval skill whose substrate is the L2 summary table. Each result IS a citation back to the underlying conversation (via conversation_id) and to the summary row (for downstream L3 consumers). The "extract mode" parameter projects different structured fields from the summary payload — still no synthesis, just projection.
semantic-search finds candidate messages and you want the structured versionextract="questions" → open questions only. extract="decisions" → decisions only. extract="quotes" → quotables only. Default extract="summary".query: str
extract: "summary" | "questions" | "decisions" | "quotes" = "summary"
limit: int = 10
domain: str | None
importance: "breakthrough" | "significant" | "routine" | None
thinking_stage: "exploring" | "crystallizing" | "refining" | "executing" | None
source: str | None
mode: "hybrid" | "fts" | "vector" = "hybrid"
Markdown with per-row citations. Every row carries [conv_id · date] and [summary_id]. Extract modes project different structured fields but never paraphrase.
what-do-i-think)[a-zA-Z0-9\s\-_.,] is enforced)