PlugWiki

recall Skill (agentmemory) for Claude Code: Search Memory

Searches agentmemory for what you decided or learned in past sessions, for anyone who asks Claude "did we ever" or "what did we do about".

Installs
14.2K
on skills.sh
Stars
30.1K
on GitHub
Files
2
besides SKILL.md
Updated
Oct 2026
last repo commit

recall is the search command of the agentmemory plugin. You type /recall with a topic, and Claude searches saved memories and session records with a mix of keyword, vector and graph search. It lists what it found, grouped by session, with the most important items first.

It triggers on phrases like "recall", "what did we do about", "did we ever" and "have we seen". Its rule is to show only what the tool returned. If nothing matches, it suggests 2 to 3 other search terms instead of guessing. It bundles no scripts and needs the agentmemory server and its MCP tools.

Good for

  • Developers returning to a project after days away
  • Anyone who asks "have we seen this error before"
  • Teams sharing one memory store across coding agents

Install recall

Claude Code

  1. 1

    Copy this prompt.

    Install the recall skill in Claude Code. Run this command:

npx -y @agentmemory/agentmemory@latest

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js 20 or newer.

This skill only works when the memory server is running and the agentmemory MCP tools are connected. The plugin route is the one the README recommends for Claude Code. The skills-only route installs all 17 skills but does not connect the MCP tools by itself. Pick one route, not both.

From the project README. Source.

What's in the skill

When it loads

Search agentmemory for past observations, sessions, and learnings about a topic using hybrid BM25 plus vector plus graph search. Use when the user says "recall", "what did we do about", "did we ever", "have we seen", or needs context from past sessions.

Sections in SKILL.md

  1. Quick start
  2. Why
  3. Workflow
  4. Anti-patterns
  5. Checklist
  6. See also
  7. Troubleshooting

Key facts

Author
rohitg00
Repository
rohitg00/agentmemory
Installs on skills.sh
14,234 as of 2026-11-02
GitHub stars
30,069 for the whole rohitg00/agentmemory repo
License
Apache-2.0 (repo and plugin.json)
Last repo commit
2026-10-29
What is inside
3 files including SKILL.md: SKILL.md, EXAMPLES.md and TROUBLESHOOTING.md. It bundles no scripts.
Part of
The agentmemory plugin (17 skills, 12 hooks, 54 MCP tools per plugin.json)
Needs
The agentmemory server running, and its memory_smart_search tool connected

What it includes

  • Slash command that searches past memories on a topic

    Takes your topic as the argument and searches agentmemory with it.

    Try: “/recall jwt refresh token rotation”

    Read · /recall [search query]

  • Hybrid search with limit, project, agent and layer filters

    Claude calls it with a limit of 10. It can be scoped by project and agent, and by layer: "memory" for saved knowledge or "observation" for captured activity.

    Network · memory_smart_search

  • Fetches full records by ID before quoting them

    Search results are compact. Before quoting full content, Claude calls the search again with the chosen IDs. A title alone is not treated as the full fact.

    Read · Expand results

  • Reads matched_count, excluded_by_budget and content_truncated

    An empty or shortened reply is not read as absent knowledge until these fields are checked.

    Read · memory_recall checks

  • Shows type, title and narrative, high importance first

    Observations are grouped by session and those with importance 7 or higher lead.

    Group by session

  • Suggests 2 to 3 other terms and stops

    If nothing matches, it says so. It must not invent an observation, a session ID or an importance score.

    No-match rule

From the official docs

Permissions

Medium risk

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • The plugin records your sessions through hooks

    The README lists captured data: user prompts (privacy-filtered), tool name, input and output, errors, and the assistant reply of each turn. Review what is stored in the viewer, and use the forget skill to remove entries. github.com

  • Installing starts a local server and runs npm code

    The setup command pulls the latest package with npx and starts a background server. On macOS and Linux it also needs curl, sh and tar for automatic runtime setup. github.com

  • Saved memories flow back into later sessions

    plugin.json says the plugin injects context into future sessions. In the hook scripts, SessionStart and PreToolUse inject context only when AGENTMEMORY_INJECT_CONTEXT=true, as the README also says. The PreCompact script asks the server for context before each compaction with a default budget of 1500 and prints it as hook output; AGENTMEMORY_PRE_COMPACT_BUDGET=0 turns that off. Our advice: avoid saving text you do not trust, because the assistant reads it back. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details
Turning on read-onlyNo read-only switch is documented for these skills.
Talks toThe server and viewer bind to loopback (127.0.0.1) by default. Compression or embeddings with a provider you configure use that provider. Local semantic search downloads an embedding model on first use if you set EMBEDDING_PROVIDER=local
Sign-in and keysA local server creates a random secret at ~/.agentmemory/secret with mode 0600 and the bundled clients read it. A remote server needs AGENTMEMORY_SECRET in your environment.
Data retentionMemories are stored on your machine in the agentmemory data folder until you delete them. The README says API keys, bearer tokens and <private> tags are stripped before storage.
Scopes

  • The Claude Code plugin registers 12 hooks that run node scripts on events such as every prompt, every tool call and the end of each turn

  • The plugin ships a .mcp.json that starts a node bridge to the local memory server

  • Setup runs npx -y @agentmemory/agentmemory@latest, writes config to ~/.agentmemory and starts a server on localhost:3111

  • If the MCP tools are missing, the skill's TROUBLESHOOTING.md tells the assistant to call the REST API on the daemon directly with a bearer secret, read from AGENTMEMORY_SECRET or, for loopback URLs, from ~/.agentmemory/.env and ~/.agentmemory/secret

Questions

What does the recall skill do?

It searches what agentmemory has stored about past sessions and saved notes, and reports only what comes back. Results are grouped by session with the high-importance ones first.

How do I install recall?

Use the install box at the top of this page. Start the agentmemory server first, then add the plugin. The skill calls search tools the server provides, so the folder alone does nothing.

What happens when it finds nothing?

It says no memories matched and suggests 2 to 3 alternative search terms. The skill forbids guessing from assumption.

Does recall need an embedding model or API key?

The README says keyless mode uses keyword (BM25) search, and that graph matches can be added when graph data exists. Local or remote embeddings add semantic search. Keyless installs disable vectors.

How is recall different from session-history?

recall searches by topic across all sessions. session-history lists recent sessions on the current project as a timeline. Both read the same stored data.

Is recall safe?

Its skill file tells Claude to call a search tool and report what comes back; it does not list a delete or save step. The setup around it is larger: the plugin adds hooks that record your sessions and runs a local server. See the safety section on this page.

Information

updated Nov 2, 2026

Repository
rohitg00/agentmemory
Folder
plugin/skills/recall

rohitg00/agentmemoryskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 11
  1. rohitg00/agentmemory README · github.com
  2. skills CLI commands · github.com
  3. recall SKILL.md · github.com
  4. session-history SKILL.md · github.com
  5. plugin hooks.json · github.com
  6. plugin .mcp.json · github.com
  7. plugin.json · github.com
  8. Shared TROUBLESHOOTING.md · github.com
  9. pre-compact.mjs · github.com
  10. TROUBLESHOOTING.md · github.com
  11. session-start.mjs · github.com

Updated Nov 2, 2026.