How to Set Up Memory Skills in OpenClaw Bazaar

Your OpenClaw agent forgets things. Every session starts fresh, context gets lost, and you end up repeating yourself. The fix is not just configuring a single memory file — it is building a proper memory architecture using skills from OpenClaw Bazaar that handle file structure, search, pruning, and compaction for you.

This guide walks through how to find, install, and configure the best memory skills from the Bazaar directory so your agent retains context across sessions without manual babysitting.

Why Memory Skills Beat Manual Configuration

Configuring memory from scratch means writing bootstrap files, setting compaction thresholds, wiring up search retrieval, and building pruning habits. Most operators get one or two pieces right and miss the rest.

Memory skills from the Bazaar directory package all of this into tested, community-rated bundles. A single skill install can set up your SOUL.md, AGENTS.md, USER.md, and MEMORY.md files with proper structure, configure pre-compaction flushing, and enable hybrid search — all in under ten minutes.

The difference is reliability. A hand-rolled memory setup works until it does not, and diagnosing why your agent forgot something is painful. A well-maintained Bazaar skill has been tested across hundreds of deployments, and the community has already found and fixed the edge cases.

The Four Files Every Memory Skill Should Create

When browsing the Bazaar for memory skills, look for ones that generate these four bootstrap files:

SOUL.md defines who your agent is. Personality, tone, and behavioral boundaries live here. This file survives compaction because it reloads from disk at every session start. A good memory skill generates a SOUL.md template with sections for identity, communication style, and hard boundaries like "never send emails without showing a draft first."

AGENTS.md contains operational rules. This is your agent's playbook — workflow conventions, retrieval protocols, and a list of things the agent should never do. The most valuable section is the Memory Protocol, which tells the agent to search memory before answering questions and to save important information immediately when it learns something new.

USER.md stores information about you. Your projects, clients, technical preferences, timezone, and communication style. This file makes every conversation feel like your agent already knows you, because it does.

MEMORY.md is the lean cheat sheet. Keep it under 100 lines. Key decisions and the reasons behind them, preferences the agent has learned from corrections, active project states, and rules from past mistakes. This file is loaded into every single API call, so every token counts.

How to Find Memory Skills in the Bazaar

Open the skills directory and filter by the "Memory" category. Sort by community rating to surface the most battle-tested options. Look for skills with 50 or more installs and a rating above 4.0 — these have been validated across diverse setups.

Read the skill description carefully. Some memory skills focus only on file architecture. Others bundle search configuration, compaction settings, and pruning automation together. The best ones do all of it.

Check the skill's requirements section. Some memory skills need a specific OpenClaw version or assume you are using a particular model provider. Make sure your setup meets the prerequisites before installing.

Installing a Memory Skill From the Bazaar

Installation is straightforward. From your OpenClaw interface, run the install command with the skill's Bazaar identifier. The skill will generate its bootstrap files in your workspace directory and add configuration entries to your agent's settings.

After installation, run /context list to verify everything loaded correctly. You should see your new SOUL.md, AGENTS.md, USER.md, and MEMORY.md files in the output. Check that none of them show "TRUNCATED" — if they do, the file is too long and the agent is working with incomplete information.

Fill in the template sections with your actual information. The skill generates placeholder content, but the value comes from your specific context. Spend fifteen minutes filling in your projects, preferences, and operational rules. This upfront investment pays for itself within the first week.

Configuring Search and Retrieval

The best memory skills from the Bazaar configure hybrid search automatically. Hybrid search combines keyword matching (finding exact word matches) with semantic embedding (finding conceptually related content even when the words differ). A typical configuration uses 70 percent vector weight and 30 percent keyword weight.

The embedding model downloads automatically the first time you use search. It runs locally on your machine with no API calls, so there is no ongoing cost for search operations.

If you have a large knowledge base — an Obsidian vault, project documentation, or hundreds of memory files — look for skills that support extra path indexing. These skills let you point the search system at additional directories beyond the default memory folder.

For operators with thousands of files, the Bazaar offers QMD-based skills that return small, relevant snippets instead of entire documents. This keeps your context window lean and avoids triggering compaction when retrieving information from large collections.

Setting Up Compaction Protection

Compaction is where most memory systems fail silently. When your conversation history gets too large, OpenClaw summarizes it to free up space. This summarization is lossy — details disappear into compressed summaries and never come back.

The critical setting is reserveTokensFloor. Memory skills from the Bazaar typically set this to 40,000 tokens, which gives the pre-compaction memory flush enough headroom to fire before the context overflows. The default value is often too tight, and a single large file read can jump past the threshold before the flush gets a chance to save anything.

The pre-compaction flush is a safety net that triggers a save-to-disk step right before compaction runs. It tells the agent: "You are about to lose context. Save anything important to your daily log file now." Good memory skills configure this automatically with appropriate thresholds and prompts.

Pruning: The Maintenance Nobody Does

Memory files grow over time. Without regular pruning, they become bloated with outdated information, duplicate entries, and content that no longer matters. Bloated memory costs more tokens per API call, reduces retrieval quality because relevant content gets buried, and slows down search.

The Bazaar has dedicated pruning skills that automate this maintenance. These skills run on a schedule — typically weekly — and analyze your memory files for outdated information, duplicates, and entries that have not been referenced recently. They generate a pruning report for your review and archive or remove stale content after your approval.

Manual pruning follows a simple schedule. Weekly, review your daily logs and promote important decisions into MEMORY.md. Monthly, scan all memory files for outdated client information, completed projects, and preferences that have changed. Quarterly, reorganize files by current relevance and split any file over 2,000 words into focused topic-specific files.

What Good Memory Looks Like in Practice

An agent without memory asks you which client you mean every time you mention "the client." It drafts generic emails. It forgets decisions you made last week.

An agent with properly configured memory skills knows your active clients by name, remembers their communication preferences, references decisions from previous sessions, and drafts personalized responses that reflect your history with each contact.

The compounding effect is the real payoff. After a month of well-maintained memory, your agent handles routine tasks with minimal input. After three months, interactions feel natural — the agent anticipates your preferences and asks fewer clarifying questions because it already knows the answers.

The operators who get the most value from memory skills are not the ones with the most complex configurations. They are the ones who spent fifteen minutes filling in their bootstrap files honestly and then maintained a simple weekly review habit. Start there, and the Bazaar skills handle the rest.

Browse the Skills Directory

Find the right skill for your workflow. The OpenClaw Bazaar skills directory has over 2,300 community-rated skills — searchable, sortable, and free to install.

Browse Skills →

Try a Pre-Built Persona

Don't want to configure everything from scratch? OpenClaw personas come pre-loaded with skills, memory templates, and workflows designed for specific roles. Compare personas →