How do I install the Search Conversations skill?

Run npx skillmds@latest add lohomi15/search-conversations in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Search Conversations safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Search Conversations?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Search Conversations free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Search Conversations?

Lohomi15 (@lohomi15) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Search Conversations by lohomi15 — Agent Skill

Name: Search Conversations
Author: Lohomi15

Search Conversations

Search previous Claude Code conversations for a keyword or topic and return matching context. Use when the user asks "where did we talk about X", "find the chat about Y", "search past conversations", or wants to recall context from a prior session.

Lohomi15 Updated Sep 17, 2026

File contents

search-conversations

Finds mentions of a keyword or phrase across every Claude Code session stored in ~/.claude/projects/*/*.jsonl and returns matching snippets with surrounding context.

When to use

User asks about a prior conversation: "where did we discuss X", "which chat was it where we talked about Y", "find the session about Z".
User wants to recover context from a past session that isn't in memory.
User references something you don't recognize and suspects it was discussed before.

How to run

Pass the query as a single argument. Quote multi-word phrases.

python3 ~/.claude/skills/search-conversations/search.py "nagarhole"
python3 ~/.claude/skills/search-conversations/search.py "coorg trip" -c 2 -n 10

Flags:

-c N — include N turns of context before/after each hit (default 1).
-n N — cap number of hits (default 20).
--chars N — truncate each snippet line to N chars (default 400).
--since YYYY-MM-DD — only include matches on or after this date.
--until YYYY-MM-DD — only include matches on or before this date.

Use date flags when the user references timing ("last week", "yesterday", "in April"). Resolve the relative date to YYYY-MM-DD before passing it in.

Output is grouped per hit: timestamp, session id (first 8 chars), project dir, a context window of [user]/[assistant] turns, and the full jsonl path.

What to do with results

Summarize the matching conversations for the user — dates, what was discussed, which session(s).
If the user wants more detail, read the jsonl path directly with the Read tool or grep it for surrounding turns.
If there are no matches, say so plainly; don't fabricate recall.
Offer to resume the best match (see below). Users usually want to jump back in, not just read a snippet.

Resuming a matched conversation

/resume is a built-in Claude Code slash command — it can't be invoked via Skill or Bash. Only the user can type it. So the flow is confirm → print command → user runs it.

After presenting matches, pick the most likely target (usually the newest hit, or the one whose snippet best matches the query). If there are multiple strong candidates, list them briefly and ask which.
Confirm with the user in one line: "Want to resume this one?"
On yes, print the full session UUID with the exact command to run:

Run /resume <full-session-uuid> to jump back in.

Guidelines:

Always print the full UUID, not the truncated 8-char prefix shown in search output. Pull it from the sessionId field or the jsonl filename.
Do NOT attempt to call /resume yourself via Skill or Bash — it will fail.

Notes

Search is case-insensitive and substring-based.
Results are sorted newest first.
The current active session is included — matches there are just "earlier in this chat".

---name: search-conversationsdescription: Search previous Claude Code conversations for a keyword or topic and return matching context. Use when the user asks "where did we talk about X", "find the chat about Y", "search past conversations", or wants to recall context from a prior session.---# search-conversationsFinds mentions of a keyword or phrase across every Claude Code session stored in `~/.claude/projects/*/*.jsonl` and returns matching snippets with surrounding context.## When to use- User asks about a prior conversation: "where did we discuss X", "which chat was it where we talked about Y", "find the session about Z".- User wants to recover context from a past session that isn't in memory.- User references something you don't recognize and suspects it was discussed before.## How to runPass the query as a single argument. Quote multi-word phrases.```bashpython3 ~/.claude/skills/search-conversations/search.py "nagarhole"python3 ~/.claude/skills/search-conversations/search.py "coorg trip" -c 2 -n 10```Flags:- `-c N` — include N turns of context before/after each hit (default 1).- `-n N` — cap number of hits (default 20).- `--chars N` — truncate each snippet line to N chars (default 400).- `--since YYYY-MM-DD` — only include matches on or after this date.- `--until YYYY-MM-DD` — only include matches on or before this date.Use date flags when the user references timing ("last week", "yesterday", "in April"). Resolve the relative date to YYYY-MM-DD before passing it in.Output is grouped per hit: timestamp, session id (first 8 chars), project dir, a context window of `[user]`/`[assistant]` turns, and the full jsonl path.## What to do with results1. Summarize the matching conversations for the user — dates, what was discussed, which session(s).2. If the user wants more detail, read the jsonl path directly with the Read tool or grep it for surrounding turns.3. If there are no matches, say so plainly; don't fabricate recall.4. **Offer to resume** the best match (see below). Users usually want to jump back in, not just read a snippet.## Resuming a matched conversation`/resume` is a built-in Claude Code slash command — it can't be invoked via Skill or Bash. Only the user can type it. So the flow is confirm → print command → user runs it.1. After presenting matches, pick the most likely target (usually the newest hit, or the one whose snippet best matches the query). If there are multiple strong candidates, list them briefly and ask which.2. Confirm with the user in one line: *"Want to resume this one?"*3. On yes, print the **full** session UUID with the exact command to run: > Run `/resume <full-session-uuid>` to jump back in.Guidelines:- Always print the full UUID, not the truncated 8-char prefix shown in search output. Pull it from the `sessionId` field or the jsonl filename.- Do NOT attempt to call `/resume` yourself via Skill or Bash — it will fail.## Notes- Search is case-insensitive and substring-based.- Results are sorted newest first.- The current active session is included — matches there are just "earlier in this chat".

Frequently asked questions