remembering-conversations
obra/episodic-memory
Forces an AI agent to search past conversation history before guessing or saying "I don't know."
What is remembering-conversations?
This skill instructs an AI agent to search historical conversation logs via an episodic-memory MCP server before answering from guesswork, saying "I don't know," or treating a topic as new. It uses a search-conversations subagent (or direct MCP search/read tools) to retrieve and synthesize relevant past discussions, decisions, and gotchas.
- Triggers a mandatory search of historical conversation memory before guessing or claiming ignorance
- Dispatches a 'search-conversations' subagent (Claude Code Task tool) or equivalent agent (Codex) to run the search
- Falls back to direct episodic-memory MCP 'search' and 'read' tools when subagent dispatch isn't available
- Reads only the top 2-5 relevant results to minimize context usage
- Synthesizes findings into a concise summary with source pointers to original conversations
How to install remembering-conversations
npx skills add https://github.com/obra/episodic-memory --skill remembering-conversations- episodic-memory MCP server/plugin installed and configured (provides search and read tools)
- For Claude Code: a 'search-conversations' subagent must be available for Task tool dispatch
- For Codex: either a 'search-conversations' agent or direct MCP tool access to episodic-memory search/read tools
How to use remembering-conversations
- 1.Recognize a trigger: you're about to say 'I don't know', guess, treat a topic as new, get stuck, or the user references past discussion ('last time', 'we discussed', 'do you remember').
- 2.Announce intent: state 'Searching past conversations for [topic].'
- 3.In Claude Code, dispatch the Task tool with subagent_type 'search-conversations' and a prompt describing the query and what you're looking for (decisions, patterns, gotchas, code examples).
- 4.In Codex, dispatch the same search-conversations agent if available; otherwise call the episodic-memory MCP 'search' tool directly.
- 5.Read the top 2-5 results using the episodic-memory MCP 'read' tool.
- 6.Synthesize the findings into a concise (200-1000 word) answer with actionable insights.
- 7.Include source pointers/citations so the user can inspect the original conversations.
- 8.Consult MCP-TOOLS.md for the full API reference if advanced usage is needed.
Use cases
- Recalling a past decision and its rationale before re-deciding it from scratch
- Finding prior solutions or workarounds when stuck on a bug that resembles a past issue
- Answering 'why did we...' or 'what do we know about...' questions from project history
- Repeating a previously-established workflow that has known gotchas or steps
- Avoiding repeated mistakes by checking if a topic was already addressed in an earlier conversation
- Users of Claude Code or Codex who want their agent to leverage prior conversation history automatically
- Teams maintaining long-running projects where decisions and context accumulate across many sessions
- Developers who want to reduce repeated debugging or re-discussion of previously solved problems
remembering-conversations FAQ
No, it supplements them. It's described as filling gaps left by other memory systems that only hold partial records, by letting the agent search actual past conversation transcripts.
Use the episodic-memory MCP tools directly: call the `search` tool, then `read` the top 2-5 results, and synthesize the findings yourself.
Don't use it to explore current codebase structure (use Grep/Read instead), for info already in the current conversation, or before you understand what you're being asked to do.
The skill claims it saves 50-100x context compared to loading raw conversation history directly.
Full instructions (SKILL.md)
Source of truth, from obra/episodic-memory.
name: remembering-conversations description: 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.
Remembering Conversations
Core principle: Search before reinventing. Searching costs nothing; reinventing or repeating mistakes costs everything.
Mandatory: Search Historical Memory
YOU MUST search historical memory for any historical search.
Announce: "Searching past conversations for [topic]."
Claude Code
Use the Task tool with subagent_type: "search-conversations":
Task tool:
description: "Search past conversations for [topic]"
prompt: "Search for [specific query or topic]. Focus on [what you're looking for - e.g., decisions, patterns, gotchas, code examples]."
subagent_type: "search-conversations"
Codex
If a search-conversations agent is available, dispatch it with the same prompt. If not, use the MCP tools directly:
- Search with the episodic-memory
searchtool - Read the top 2-5 results with the episodic-memory
readtool - Synthesize findings in your response
- Include source pointers so the user can inspect the original conversations
The search workflow will:
- Search with the
searchtool - Read top 2-5 results with the
readtool - Synthesize findings (200-1000 words)
- Return actionable insights + sources
Saves 50-100x context vs. loading raw conversations.
When to Use
Use this whenever the current task would benefit from information you may have learned before, even if the user did not explicitly ask you to search.
When past experience may help:
- You need to recall decisions, rationale, patterns, solutions, pitfalls, or project context from earlier work
- A task resembles something you've solved, debugged, reviewed, released, or planned before
- You need to repeat a workflow or process that may have prior gotchas or established steps
When you're stuck:
- You've investigated a problem and can't find the solution
- Facing a complex problem without obvious solution in current code
- Need to follow an unfamiliar workflow or process
When historical signals are present:
- User says "last time", "before", "we discussed", "you implemented"
- User asks "why did we...", "what was the reason..."
- User says "do you remember...", "what do we know about..."
Before answering from uncertainty:
- Before guessing from memory or saying "I don't know" about something that may have been learned in a past conversation, search memory unless the current conversation already answers it
Don't search first:
- For current codebase structure (use Grep/Read to explore first)
- For info in current conversation
- Before understanding what you're being asked to do
Direct MCP Tool Access
Use these directly when a search agent is unavailable or the current harness does not support agent dispatch:
mcp__plugin_episodic-memory_episodic-memory__searchmcp__plugin_episodic-memory_episodic-memory__read
When using MCP tools directly, keep context small: search first, then read only the top 2-5 relevant conversations or line ranges.
See MCP-TOOLS.md for complete API reference if needed for advanced usage.
Related skills
More from obra/episodic-memory and the wider catalog.
find-skills
Discover and install agent skills to extend your coding agent's capabilities on demand
frontend-design
Build visually distinctive UI with opinionated aesthetic direction, typography, and layout choices that avoid templated defaults.
vercel-react-best-practices
70 React/Next.js performance rules from Vercel Engineering, prioritized by impact for writing, reviewing, and refactoring code.
agent-browser
Fast browser automation CLI for AI agents — navigate, click, scrape, screenshot, and test via Chrome CDP
web-design-guidelines
Review UI code against Web Interface Guidelines for accessibility, UX, and design best practices
finetuning
Fine-tune models on Azure AI Foundry with SFT, DPO, or RFT training methods.