菜单
Synthesize the user's views on a topic across 10+ conversations. Modes synthesize (default) or precedent. Use when user is forming an opinion and may have already concluded something.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
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 | what-do-i-think |
| description | Synthesize the user's views on a topic across 10+ conversations. Modes synthesize (default) or precedent. Use when user is forming an opinion and may have already concluded something. |
| layer | L2 |
| reads | ["L2.summaries","L1.embeddings"] |
| writes | [] |
| citations | required |
| determinism | temporal |
| allowed-tools | mcp__my-brain__what_do_i_think |
L2 synthesis that surfaces what the user has already decided/concluded about a topic before they re-ask. Deduplicates by 80-char lowercase prefix. Orders by importance (breakthrough > significant > routine). Every decision/question/quote carries a citation.
Precedent mode is a narrower projection: returns the closest past matches for a situation, filtered to exclude "none identified" decisions.
precedent when user cites a specific situation ("I'm facing a decision like Z")topic: str
mode: "synthesize" | "precedent" = "synthesize"
Synthesize mode:
## What you think about: {topic}
### Top 5 summaries
{icon} **{title}** [conv_id · date]
> {summary_300ch}
### Key decisions (dedup by 80ch prefix, top 10)
- {decision} [conv_id · date]
### Still open (top 8)
- {question} [conv_id · date]
### Authentic quotes (top 5)
> "{quote}" [conv_id · date]
Precedent mode: top 10 closest matches, each row cites one summary with decision field.
No summaries → Lance + keyword search on L0. Appends: "Running without summaries. For richer synthesis: brain-mcp summarize"