echo-sleuth

echo-sleuth

Mine past Claude Code conversations and manage the knowledge lifecycle — search sessions, extract lessons, audit memory staleness, prune token waste.

What it does

Echo Sleuth analyzes your Claude Code session history (~/.claude/projects/) and memory files to help you:

• Search past sessions — find conversations by keyword, date, or file
• Summarize recent work — get a quick recap of what happened across sessions
• Build timelines — chronological project history combining sessions and git commits
• Extract lessons — surface values, patterns, mistakes, and decisions worth remembering
• Audit memories — find stale, broken, or wasteful memories across all projects
• Extract knowledge — distill conversations into durable memories, CLAUDE.md rules, or human notes
• Prune waste — interactively clean up memories that silently eat tokens and money

Part of the xiaolai plugin marketplace.

Installation

Add the marketplace (once):

/plugin marketplace add xiaolai/claude-plugin-marketplace

Then install:

/plugin install echo-sleuth@xiaolai

Install fails with "Plugin not found in marketplace 'xiaolai'"? Your local marketplace clone is stale. Run claude plugin marketplace update xiaolai and retry — plugin install does not auto-refresh.

Scope	Command	Effect
User (default)	/plugin install echo-sleuth@xiaolai	Available in all your projects
Project	/plugin install echo-sleuth@xiaolai --scope project	Shared with team via .claude/settings.json
Local	/plugin install echo-sleuth@xiaolai --scope local	Only you, only this repo

Commands

When installed as a plugin, commands appear as /echo-sleuth:<command> (e.g. /echo-sleuth:recall).

Conversation Mining

/recall <topic> [--scope current|all] [--limit N] [--lite]

Search past conversations for a topic, decision, or mistake.

/echo-sleuth:recall "why did we choose SQLite"
/echo-sleuth:recall "authentication bug" --scope all
/echo-sleuth:recall "CI setup" --limit 20
/echo-sleuth:recall vitepress --lite

• Searches session summaries, first prompts, and message content
• Auto-detects focus: decision archaeology ("why did we..."), mistake hunting ("what went wrong..."), or general search
• Default scope: current project. Use --scope all to search across all projects.
• --lite skips agent dispatch and synthesis. The slash command runs recall-lite.sh and returns its raw output verbatim — minimum-tokens mode for when you hit billing limits or just want raw evidence. See Lite mode and API usage below.

/recap [N-sessions|duration] [--detail low|medium|high]

Summarize recent sessions.

/echo-sleuth:recap
/echo-sleuth:recap 10
/echo-sleuth:recap 7d --detail high

• Default: last 5 sessions, medium detail
• Accepts session count (10) or duration (3d, 1w)
• Detail levels: low (one-liners), medium (paragraph per session), high (full analysis)

/timeline [--limit N] [--since YYYY-MM-DD]

Chronological project history combining Claude sessions and git commits.

/echo-sleuth:timeline
/echo-sleuth:timeline --since 2026-03-01 --limit 50

• Merges session timestamps with git commit history
• Shows what work happened when, in what order
• Requires git for commit correlation

/lessons [topic] [--scope current|all] [--category decisions|mistakes|patterns|all]

Extract accumulated wisdom from past sessions.

/echo-sleuth:lessons
/echo-sleuth:lessons "database" --category decisions
/echo-sleuth:lessons --scope all --category mistakes

• Surveys all sessions, samples the highest-signal ones (longest, most errors, most recent)
• Categories: learned values, decisions, mistakes, patterns, tool preferences, architecture insights, cost/efficiency
• Cross-references with git history when available

Memory Management

/dashboard

Global overview of Claude Code memories across all projects.

/echo-sleuth:dashboard

Shows:

• Summary stats — how many projects have memories, total files, estimated token load per conversation
• Staleness alerts — memories scored > 50 on the staleness scale, with age, type, and recommended action
• Top token consumers — which projects' memories cost the most tokens

Use this as the entry point to understand your memory landscape before auditing or pruning.

/audit [project] [--deep]

Audit memory staleness — quick heuristic scan or deep content verification.

/echo-sleuth:audit
/echo-sleuth:audit pixel-office
/echo-sleuth:audit --deep
/echo-sleuth:audit pixel-office --deep