Memory Recall

Memory Recall

Load and intelligently summarise stored memory so a session can pick up exactly where the last one ended.

Works in claude.ai (reads via memory_user_edits view) and Claude Code (reads ~/.claude/memory.md). Detected automatically.

---

Step 0 — Detect environment

| Signal | Environment | Load | |--------|-------------|------| | memory_user_edits tool is available | claude.ai | references/claude-ai.md | | Not available / running in terminal | Claude Code | references/claude-code.md |

Read the relevant reference file before proceeding.

---

Core Protocol

Step 1 — Read the full memory store

Use the environment-appropriate read command (see references) to fetch all stored entries.

Step 2 — Sort entries by relevance to the current request

Memory entries fall into priority tiers:

Tier 1 — Always surface:

• CURRENT STATE SNAPSHOT (what's live, what's imminent)
• Anything tagged "PENDING" or "next" in entries

Tier 2 — Surface if the user asked for context:

• Project deliverables, decisions, frameworks
• Recent event details, dates, links

Tier 3 — Surface only if directly relevant to the immediate query:

• Historical context, completed work, design tokens
• Personal/operating preferences (PWP, communication style)

Step 3 — Present the recall

Match the format to the trigger:

Proactive recall (start of session, no specific question):

Picking up where we left off:

Current state — [from CURRENT STATE SNAPSHOT entry, paraphrased]

Pending items:
→ [item 1]
→ [item 2]
→ [item 3]

Active projects: [list]

Say "show me the full memory map" to see everything stored.

On-demand recall ("what do you remember?"):

Memory map — [N] entries stored

CURRENT STATE
[paraphrase]

PROJECTS
→ [project 1]: [one-line status]
→ [project 2]: [one-line status]

DELIVERABLES
[count] documents, [count] emails, [count] events

PREFERENCES
[note operating style, tools, communication preferences]

Specific recall ("what do you remember about X?"):

Filter to entries matching X. Present as a focused brief, not a dump.

---

Hard rules

• Never read sensitive entries aloud unless the user explicitly asked about that topic
• Never invent or extrapolate beyond what's stored — if memory is silent on a topic, say so
• If memory is empty, say it cleanly: "No stored memory yet. I'll start tracking once we build something together."
• Don't surface low-tier entries proactively — they create noise
• Always offer the path to "show me everything" rather than dumping all entries by default

---

Pairs with

• memory-map — writes new memory at session close
• memory-prune — cleans stale entries when memory gets cluttered

Together: write → recall → prune. The full memory loop.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: shandar
• Source: shandar/memory-map
• License: MIT
• Homepage: https://affordance.design

Install and usage instructions live in the source repository linked above.