Atomic Memory Manager

Recall and write workspace memory for OpenClaw agents using MEMORY.md, memory/YYYY-MM-DD.md, memory_search, and memory_get. Use when the user asks about prio...

Security Scan

The skill's instructions, requirements, and behavior are internally consistent with a workspace memory manager and do not request unrelated credentials or installs.

Purpose & Capability

Name/description match the instructions: it explains when to retrieve and when to write workspace memory files (MEMORY.md and memory/YYYY-MM-DD.md) and references memory_search/memory_get. It does not request unrelated binaries, env vars, or config paths.

Instruction Scope

Instructions stay within the stated purpose: use memory_search and memory_get for retrieval and write small durable notes to MEMORY.md or daily files. They explicitly avoid secrets unless asked and ask the agent to cite file paths. Note: this is purely procedural (no implementation included) and assumes platform-provided memory_search/memory_get primitives; it also relies on the agent not being asked to store secrets — if a user asks the agent to 'remember' sensitive data, those instructions could cause sensitive data to be written to workspace files.

Install Mechanism

Instruction-only skill with no install spec or downloaded code; nothing is written to disk by the skill itself beyond what the agent is instructed to do at runtime.

Credentials

The skill requests no environment variables, credentials, or config paths — proportional to its goal of managing local workspace memory.

Persistence & Privilege

always:false and user-invocable:true. The skill instructs the agent to write files in the workspace (expected for a memory manager) but does not request global system changes or other skills' configurations.

Assessment

This skill is coherent and low-risk relative to its stated purpose. It is instruction-only, so it won't install code, but if invoked the agent will read and write memory files in your workspace (MEMORY.md and memory/YYYY-MM-DD.md). Before using it: (1) avoid asking the agent to 'remember' secrets (the instructions say 'no secrets unless asked' but will follow user requests), (2) confirm how memory_search/memory_get are implemented on your platform and whether those memory files are backed up or visible to others, and (3) periodically review the MEMORY.md and memory/ files to remove any sensitive or stale entries.

Memory Management

Retrieve first. Write only what should persist.

Core model

• MEMORY.md — curated long-term memory
• memory/YYYY-MM-DD.md — daily notes and running context
• memory_search — find likely snippets
• memory_get — read exact lines needed

Retrieval

When user asks about prior work, decisions, dates, people, preferences, or todos:

1. memory_search with a tight, specific query.
2. memory_get for the needed lines only.
3. Answer from retrieved text, not guesswork.
4. If results are weak or empty, say you checked.

Good queries: omar internship goal fall 2026, notion workspace page id Bad queries: whole paragraphs, vague terms like memory

Writing

Write when user says "remember this," states a durable preference, or makes a decision that matters later.

Destination	What goes there
MEMORY.md	Standing preferences, durable facts, recurring workflows, important long-term decisions
memory/YYYY-MM-DD.md	Day-to-day notes, progress, one-off events, session context

Rules:

• smallest useful note
• concrete facts over commentary
• no secrets unless asked
• do not promise memory unless written

When not to use

Skip retrieval when the answer is in the current message or task is current-turn only. Skip writing when the fact is trivial, belongs in project docs, or is temporary scratch.

Answering rules

• Cite file path/line when useful
• Separate memory facts from inference
• Flag stale memory
• If unavailable, say so