菜单
Summarize what's known about a project, area, or topic from memini. Use this skill when the user asks "catch me up on X", "what's the state of Y", or "summarize what we know about Z". Builds on recall with a larger limit and tiered grouping. The MCP tool is `memory_recall` on the bundled `memini` server.
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。
Import older Claude Code sessions into memini. Requires the local memini binary.
Run read-only memini store and namespace diagnostics. Requires the local memini binary.
Delete a specific memini memory safely. Use when the user asks to forget, remove, or permanently delete stored memory.
Show, set, clear, or migrate memini namespace pins with provenance. Use when project memories resolve to the wrong namespace.
Pin or unpin a memini memory while preserving its existing tags. Use when the user wants a memory in every briefing.
Show memini connection health, effective settings, namespace provenance, and read set. Use for configuration or recall diagnostics.
基于 SOC 职业分类
| name | recap |
| description | Summarize what's known about a project, area, or topic from memini. Use this skill when the user asks "catch me up on X", "what's the state of Y", or "summarize what we know about Z". Builds on recall with a larger limit and tiered grouping. The MCP tool is `memory_recall` on the bundled `memini` server. |
memini is a memory service. The fastest recap is the memory_briefing tool: it
returns pinned context, durable facts, how-to procedures, and recent activity
for the namespace in one query-less call, already grouped — use it for a
"where are we" overview. Its scope_header line spells out the ancestor chain
you inherit from (e.g. Scope: acme/phoenix/api ← acme/phoenix(3) ← acme(4));
read it instead of guessing where knowledge lives. scope widens the call:
project (just this namespace), full (default: project plus inherited
ancestor/personal/link context), or everywhere (full plus a rollup of nested
sub-projects) — use everywhere when the user wants a recap spanning
sub-projects too. For a topic-specific recap, call memory_recall with a
broad query and a higher limit (10–20) and group the results by tier.
Call memory_recall once with a broad query and limit: 20. Don't make
multiple small queries — memini's RRF ranking already handles breadth. When the
recap is scoped to a known subject rather than a fuzzy topic ("catch me up on
the deployment runbook"), use memory_list with tags / metadata filters
instead — it browses that bucket without a query (see the recall skill).
Group the response into:
semantic) — durable facts about the
project that should constrain current work.procedural) — how-to steps the user will need.episodic) — what happened recently, ordered
by recency if possible.Don't include working-tier memories in a recap — those are transient by design.
When you attribute a decision or fact to memory, quote the stored content verbatim rather than paraphrasing it into something firmer than it says.
If the result set is empty, say so. Don't fabricate a recap from general
knowledge of the codebase. If memini is unreachable, report that and
suggest memini doctor rather than guessing.