---

name: conversation-session-memory-recap
description: Use at the start of a returning session (gap greater than four hours, or on a /resume command) to restate the matter context in two to three lines before continuing work. This is a core behavioral skill that prevents the user from having to re-explain context they already provided. Triggers on session re-entry signals and explicit resume requests. Works with long-thread compression when the prior session was lengthy.
license: MIT
metadata:
id: conversation.session-memory-recap
category: conversation
priority: P1
intent: [core, session memory, context recap, resume, matter continuity]
related: [conversation-long-thread-compression, conversation-professional-b2b, conversation-uncertainty-language]
source: Louis — HAQQ Legal AI (github.com/sboghossian/mini-claude-for-legal)
version: "1.0"

Session Memory Recap

When this applies

Activate when a returning user re-opens or resumes a legal matter session after:

• A gap of more than four hours since the last turn in the conversation.
• An explicit /resume command or equivalent "pick up where we left off" signal.
• A platform-level session boundary (e.g., a new login, a new chat window reopened from a saved matter).

Do not activate the recap when:

• The session is continuous (less than 4 hours since the last turn) — the user does not need a reminder in the same sitting.
• The user opens with a fresh, unrelated request — they have moved on; recap would be irrelevant and intrusive.
• It is the first session on a low-stakes consumer matter (e.g., a quick NDA intake that will complete in one sitting) — a recap at the start of a new session on a trivial prior query is unnecessary overhead.
• The prior session produced nothing material — if the last session was a short exploratory question with no decisions or documents, skip the recap.

Behavior

Recap pattern

Deliver the recap before any other action. Two to three lines maximum. Format:

Welcome back. We were [task] for [party/matter description], governed by [jurisdiction/law], with [key term or parameter]. Last turn: [one-sentence summary of where we stopped]. Want to pick up there?

The recap ends with a simple binary: continue where we left off, or start something new. Do not ask open-ended questions in the recap — keep the cognitive load minimal.

Example recaps

NDA matter:

Welcome back. We were drafting the mutual NDA for Acme Ltd (DIFC) x XYZ Tech (UAE LLC) for a software licensing evaluation, governed by DIFC law, 3-year confidentiality term. Last turn, you asked me to add a non-solicitation carve-out to Clause 6. Want to pick up there?

Employment contract matter:

Welcome back. We were drafting an employment contract for a software engineer at your DIFC entity — AED 25,000/month, 1-year fixed-term, 3-month probation. Last turn, we were reviewing the non-compete clause and you had a question about the 1-year post-exit scope. Ready to continue?

Litigation brief matter:

Welcome back. We were drafting a Statement of Claim for filing in the DIFC Courts — Globex Ltd v. Acme Group LLC, breach of MSA, claimed amount AED 2.1 million. Last turn, the damages calculation section was open. Pick up there?

Sources for the recap

Pull from the following, in order of preference:

1. Matter metadata stored in the eFirm / matter management system (if available via integration): client name, matter number, document type, status.
2. Compressed context block (if [[conversation-long-thread-compression]] was run in the prior session): the block contains the structured matter state.
3. Last 3 turns of the prior conversation: extract the most recent task, the most recent decision, and any pending item explicitly mentioned.
4. Any pending TODOs noted explicitly in prior turns (e.g., "I'll draft Clause 7 next").

If none of these sources are available (fully fresh context), acknowledge the gap briefly: "I don't have context from your prior session — can you briefly remind me what we were working on?"

Long-thread handling

If the prior session was long (greater than approximately 50 turns) and a compression block exists, use the compression block as the primary source for the recap. The recap should still be 2–3 lines (not the full compression block) — the compression block is background context; the recap is the opening line to the user.

If the session was long but no compression was run, trigger [[conversation-long-thread-compression]] silently before delivering the recap, then recap from the resulting compressed state.

What to include in the recap

Always include	Include if material	Omit
Document type and matter parties	Jurisdiction / governing law	Verbose explanation of prior reasoning
Where the last session stopped	Key term or open issue	Intermediate drafts already superseded
Simple CTA to continue or start fresh	Deadline if approaching	All prior exchanges in full

Do not

• Write a long summary of everything that happened in the prior session — the recap is an opening line, not a briefing memo. If the user needs a full briefing, they can request a matter summary separately.
• Run the recap for a continuous session — it feels patronizing in-session.
• Guess at matter details that are not clearly supported by the prior context — if uncertain, ask one focused question rather than asserting a potentially wrong recap.
• Omit the CTA — always end with a clear action option (continue / start fresh).

Edge cases

• Multiple active matters: if the user has multiple open matters and it is unclear which one to resume, list the two or three most recent and ask which one to pick up.
• Matter transferred to a new session by sharing: if another user (e.g., a colleague) is picking up a matter started by a different user, surface the matter state without assuming the new user has the same context or authority level.
• User resumes with a correction: if the user's first turn after resuming corrects the recap ("actually we decided on 2 years, not 3"), accept the correction immediately and update the matter state.

Related skills

• [[conversation-long-thread-compression]]
• [[conversation-professional-b2b]]
• [[conversation-uncertainty-language]]