Guide · Agent skill
How to use the Recall Past Memories agent skill (recall)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The recall skill lets an AI coding agent search a persistent memory store for what was done, decided or learned in earlier sessions, such as why a technical choice was made. It is for developers and teams who already use the agentmemory tool; after this guide you will know how to install the skill, trigger it with plain requests, and read its results without mistaking an empty search for missing history.
Recall Past Memoriesrohitg00/agentmemory · Apache-2.0 Download the skill Quick start
- 1Get the ZIP. Download the recall skill ZIP from the directory page.
- 2Unzip the skill. Unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder.
- 3Or upload it in the Claude apps. Go to Settings → Capabilities → Skills and upload the ZIP.
- 4Confirm agentmemory is connected. Make sure the agentmemory tool (rohitg00/agentmemory) is installed so memory_smart_search and memory_recall are available.
- 5Ask a recall question. Start a request with a trigger phrase such as "what did we do about" followed by your topic.
- 6Review the grouped results. Read the results by session, starting with high-importance observations.
What is an agent skill, and what does recall make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches. The recall skill tells the agent to search agentmemory for past observations, sessions and learnings about a topic, using hybrid BM25, vector and graph search.
Once installed, asking something like "what did we do about refresh tokens?" makes the agent search the memory store for matching observations, decisions and session notes. The skill groups results by session, leads with the most important ones, and quotes only what the tool returned. If nothing matches, it suggests alternative search terms instead of guessing.
Install the recall skill
You can get the skill in two ways. Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder to limit it to that project. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
The skill's files are SKILL.md, EXAMPLES.md and TROUBLESHOOTING.md. Restart or reload your agent session if the skill does not appear.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into .claude/skills/ inside the repository
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check that the agentmemory tool is available
The skill does not store anything itself. It only calls two agentmemory functions, memory_smart_search and memory_recall, so the agentmemory tool (rohitg00/agentmemory) must be installed and connected. If it is not, the skill reports the search function as unavailable.
If you see "MCP tool not available", follow the steps in TROUBLESHOOTING.md. In Claude Code, the source suggests inspecting /plugin list and /mcp.
memory_smart_search { "query": "jwt refresh token rotation", "limit": 10 }Trigger the skill with natural requests
You do not need a special command. The skill is triggered by phrases such as "recall", "what did we do about", "did we ever", "have we seen", or any request for context from past sessions. The text after the trigger becomes the search query.
Be specific about the topic. A query like "what did we do about the rate limiter?" searches for rate limiter history, while a vague request returns broader and noisier results.
- Did we ever decide how to cache sessions?
- What did we do about refresh tokens?
- Recall anything about the rate limiter.
- Have we seen this 429 error before?
Read the results
Results come back in sessions, with a type, title, importance score and session id. The skill leads with observations of importance 7 or higher. It may also expand compact results to show the full narrative and provenance, when the tool returns them.
Treat an empty answer as "no matching memory", not proof that something never happened. Check the search terms and try the alternatives the agent suggests.
- Importance 7 and above are shown first
- Each observation is shown with its session id as returned
- Scope a search with a project or agent filter when the tool supports it
[importance 8] decision · "Rotate refresh tokens on every use" (session 7f3a9c21)Safety notes: scripts and network
The source states that this skill runs no scripts and uses no network access. Its only dependency is the agentmemory tool's search functions. Results are only as accurate as what was saved earlier, so verify important decisions against your own records before acting on them.
- Runs scripts: no
- Uses network: no (the agentmemory tool itself is a separate install)
- Not suitable for guaranteed-accurate records from outside sources
Tips for better recall results
Use the specific technical terms your team used when saving the memory, such as a library or function name. If the first search is empty, ask the agent for alternative terms, or try a related phrase like "session expiry" or "auth rotation".
- Name the component or library in the request
- Ask for the decision itself, such as "how did we decide to..."
- Use the suggested alternative terms rather than rephrasing from memory
What people use it for
- Finding why a caching or storage decision was made in an earlier session
- Checking whether a bug or error has been seen and fixed before
- Reviewing how a team handled a rate limiter or authentication design
- Getting context from past sessions before picking up unfinished work
Frequently asked questions
How do I install the recall agent skill?
Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need before using the recall skill?
You need the agentmemory tool (rohitg00/agentmemory) installed, with its memory_smart_search and memory_recall functions available. The skill only searches that store and reports it unavailable without it.
Is the recall skill free to use?
The source does not state pricing or licence terms beyond the project's Apache-2.0 licence, which is listed for this skill. Check the licence file in the repository for the full terms.
Does the recall skill run scripts or use the internet?
According to the source, it runs no scripts and uses no network access. It relies on the agentmemory search functions.
Why does recall say no memories matched?
The memory store has no saved observations that match your query. The skill is designed to suggest two or three alternative search terms rather than guess, so try those terms or a more specific phrase.
How does recall compare with just asking the agent from memory?
Recall searches a saved store and quotes only what it returns, which avoids invented history. A general chat answer has no access to your past sessions unless the memory tool supplies them.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Recall is a good fit for developers and teams who already keep context in agentmemory and want accurate lookups of past decisions without the agent inventing history. If you have not installed agentmemory, or you need guaranteed-accurate records from outside sources, choose a different approach, since the skill can only return what was saved earlier.
Download the skill