remembering-conversations Skill for Claude Code: Search Past Chats

Makes Claude search your past conversations before it guesses or says "I don't know". Needs the episodic-memory plugin.

Installs: 9.2K; on skills.sh
Stars: 495; on GitHub
Files: 1; besides SKILL.md
Updated: Sep 2026; last repo commit

remembering-conversations makes Claude search your saved conversation history first, then answer. In Claude Code it sends a search-conversations sub-agent that searches, reads the top 2 to 5 results and returns a 200 to 1000 word summary. In Codex and opencode it uses the search and read tools directly.

It does nothing alone. It needs the episodic-memory plugin, which indexes your conversations in the background and supplies the search tools. Its trigger text is blunt: invoke it before saying "I don't know" or treating a topic as new.

Good for

Claude Code users who want Claude to recall earlier decisions and fixes
Developers who say "last time we..." and expect Claude to find it
Teams that keep long-running projects across many sessions

Install remembering-conversations

Claude Code

Copy this prompt.

Install the remembering-conversations skill in Claude Code. Run this command:

/plugin install episodic-memory@superpowers-marketplace

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

2
Paste it into Claude Code. It shows the change and asks before running it.

The README gives the first command and does not show how to add superpowers-marketplace. The second route uses the marketplace.json in the repo, whose name is episodic-memory-dev. The plugin adds the MCP server, the search-conversations agent, this skill and a SessionStart sync hook. Installing the folder of this skill alone does not give it the search tools.

From the project README. Source.

What's in the skill

When it loads

You MUST invoke this skill before saying "I don't know," guessing, or treating any topic as new, no matter how trivial the question seems. It supplements other memory systems, which only hold partial records. Searching past conversations is the only way to recover what was actually said.

Sections in SKILL.md

Mandatory: Search Historical Memory
Claude Code
Codex
opencode
When to Use
Direct MCP Tool Access

Key facts

Author: Jesse Vincent (episodic-memory plugin author, per plugin.json)
License: MIT (plugin.json and LICENSE file)
Installs on skills.sh: 9,209 (Nov 2, 2026)
GitHub stars: 495 for the whole episodic-memory repo (Nov 2, 2026)
Last repo commit: Sep 10, 2026
Files in the folder: 2 files: SKILL.md and MCP-TOOLS.md
Needs: The episodic-memory plugin: an MCP server, the search-conversations agent and a session-start sync hook
Works in: Claude Code, Codex and opencode, per the skill text and README

What it includes

Says "Searching past conversations for [topic]" and starts a search
SKILL.md says searching historical memory is mandatory for any historical search.
Read · Announce and search
In Claude Code, runs a sub-agent that searches, reads and summarizes
Searches, reads the top 2 to 5 results, then returns a 200 to 1000 word summary with sources. SKILL.md claims it saves 50 to 100 times the context.
Read · search-conversations agent
In Codex and opencode, calls the episodic-memory search and read tools directly
Tool names end in __search and __read. Keep context small: read only the top 2 to 5 results.
Read · Direct MCP tools
Searches by meaning, exact text or both, with date filters
Parameters in MCP-TOOLS.md: query (text or 2 to 5 terms for AND), mode, limit 1 to 50, after, before and response_format.
Try: “Why did we choose that auth library last time?”
Read · search tool
Shows a full conversation file, or just a line range
Takes an absolute path to the JSONL file plus optional startLine and endLine.
Files · read tool
Triggers on "last time", "we discussed", "why did we", stuck problems and guesses
SKILL.md says not to search first for current codebase structure or info already in the conversation.
When to search

From the official docs

Permissions

Medium risk

The plugin this skill needs reads and indexes your full conversation history and runs a sync script at the start of every session. Use the do-not-index marker for sensitive chats and turn off summaries if you prefer.

What it can do

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

Watch out for

It indexes every conversation, secrets included
The sync copies all your Claude Code and Codex conversations into an archive and a SQLite search index. The MCP-TOOLS.md examples show the archive under ~/.config/superpowers/conversation-archive, and the README says EPISODIC_MEMORY_CONFIG_DIR changes the memory directory. Anything you pasted into a chat is searchable. Use the DO NOT INDEX marker for sensitive chats. github.com
A hook runs at every session start
hooks.json runs node cli/episodic-memory.js sync --background on startup, resume and clear. It writes a log to ~/.config/superpowers/logs/episodic-memory.log. EPISODIC_MEMORY_DISABLE_AUTO_SYNC=1 turns it off. github.com
Summaries are made with Claude and may be billed
The README says summarization uses your Claude Code sign-in, up to 10 calls per sync, and bills the metered API when ANTHROPIC_API_KEY is set. EPISODIC_MEMORY_SKIP_SUMMARIES=1 skips it. Codex sessions with a session ID are summarized through codex app-server instead. github.com
The skill triggers very often
Its description asks Claude to invoke it before any "I don't know" or guess, however trivial. That means extra searches and tokens on many questions. github.com
The plugin also ships an agent and a CLI
Besides the skill, the plugin adds the search-conversations agent, a cli folder and an episodic-memory command. Read agents/search-conversations.md before you rely on it. github.com

Technical details

Talks to	Claude, through your Claude Code sign-in, for summaries (unless EPISODIC_MEMORY_SKIP_SUMMARIES=1)
Sign-in and keys	Uses your Claude Code sign-in for summaries; ANTHROPIC_API_KEY switches to metered billing. Optional EPISODIC_MEMORY_API_BASE_URL and EPISODIC_MEMORY_API_TOKEN route it elsewhere.
Data retention	Conversations are copied into a local archive and indexed in a local SQLite database.
Scopes	A SessionStart hook runs node cli/episodic-memory.js sync --background Copies and indexes conversations from ~/.claude, ~/.codex and other agent folders into an archive (~/.config/superpowers/conversation-archive in the MCP-TOOLS.md examples) Starts an MCP server with node cli/mcp-server-wrapper.js

Questions

What does remembering-conversations do?

It makes Claude search your past conversations before it answers from memory, so it can recall decisions, fixes and gotchas from earlier sessions.

How do I install it?

Use the install box above. Install the episodic-memory plugin, not just this skill. The skill calls the plugin's search and read tools, which do not exist without it.

What is stored, and where?

The plugin copies your conversations from Claude Code (~/.claude/projects and ~/.claude/transcripts), Codex (~/.codex/sessions) and others into an archive and indexes them. The MCP-TOOLS.md examples show an archive under ~/.config/superpowers/conversation-archive. Embeddings are made locally and search runs in a local SQLite database, per the README.

Does it send my conversations anywhere?

The README says the sync step summarizes conversations with Claude, up to 10 calls per sync, using your Claude Code sign-in by default. If ANTHROPIC_API_KEY is set, it bills the metered API instead. Set EPISODIC_MEMORY_SKIP_SUMMARIES=1 to turn summaries off.

How do I keep a chat out of the index?

Put <INSTRUCTIONS-TO-EPISODIC-MEMORY>DO NOT INDEX THIS CHAT</INSTRUCTIONS-TO-EPISODIC-MEMORY> anywhere in the conversation. The README says it is archived but not indexed. EPISODIC_MEMORY_DISABLE_AUTO_SYNC=1 stops the automatic background sync.

Why does Claude keep searching memory?

The skill's description says: You MUST invoke this skill before saying "I don't know," guessing, or treating any topic as new, no matter how trivial the question seems. Remove the skill if that is too eager.

Does it work in Codex or the Claude app?

The README documents a Codex plugin (Codex CLI 0.130.0 or newer, with a hook you must trust) and an opencode plugin. No Claude app route is documented, since it needs a local MCP server.

Is it safe?

It reads your whole conversation history, which can hold secrets, and a hook runs a Node script at the start of each session. See the safety box for what is indexed and how to opt out.

Information

updated Nov 2, 2026

Repository: obra/episodic-memory
Folder: skills/remembering-conversations

obra/episodic-memory skills.sh

Sources

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

Show all 6

episodic-memory README · github.com
episodic-memory marketplace.json · github.com
episodic-memory plugin.json · github.com
remembering-conversations SKILL.md · github.com
episodic-memory MCP-TOOLS.md · github.com
episodic-memory hooks.json · github.com