Skills
skills/lev-os/agents/lev-find

lev-find

SKILL.md

lev-find (Backend for lev get)

lev find remains a compatibility alias. Prefer lev get.

Overview

lev get orchestrates semantic search across multiple indexes (codebase, docs, ideas, sessions, tasks) and fuses results using Reciprocal Rank Fusion (RRF) for comprehensive discovery.

When to Use

Use lev get when:

  • Cross-domain search - Finding related code, docs, and discussions
  • Architecture discovery - Locating design decisions and patterns
  • Context gathering - Understanding how something works across the codebase
  • Task awareness - Finding related work and TODOs

Use Grep/Glob when:

  • Exact matches - Searching for specific strings or patterns
  • Known locations - You know approximately where to look
  • File listing - Just need to find files by name

Basic Usage

# Search everywhere (auto-selects relevant indexes)
lev get "error handling"

# Search with explicit indexes
lev get "auth" --indexes codebase,documentation

# Include tasks in search
lev get "TODO authentication" --indexes codebase,tasks

External Search Backends

MANDATE: All research/find queries MUST load and use exa + deep-research + valyu for comprehensive discovery.

Exa (Neural Web + GitHub Search)

Exa provides semantic web search and GitHub code discovery.

Setup:

export EXA_API_KEY="your-key-here"

Usage:

# Search web for docs/examples
lev get "react authentication patterns" --scope=web

# Search GitHub repos
lev get "voice assistant typescript" --scope=github

# Combined: local code + web
lev get "deployment strategies" --scope=code,web

Valyu (Recursive Deep Research)

Valyu provides turn-based iterative refinement with confidence scoring (1-10 turns).

Usage:

# Recursive research with automatic refinement
valyu research "authentication best practices 2026" \
  --turns 5 \
  --threshold 0.85 \
  --strategy balanced

# Research strategies:
# - breadth: 20 sources, broad coverage
# - depth: 10 sources, comprehensive understanding
# - balanced: 15 sources, general-purpose (default)

# Deep research (mandate exa + deep-research + valyu)
lev get "authentication patterns" --scope=research

How it works:

  1. Turn 1: Initial search + AI answer with confidence scoring
  2. AI suggests refined query based on knowledge gaps
  3. Turns 2-N: Iterative deepening until confidence >= threshold
  4. Final synthesis across all turns

Mandatory Loading:

  • --scope=research → MUST load exa + deep-research + valyu
  • Any deep research → Use valyu for iterative refinement
  • Any find/research skill invocation → Load all backends

Config:

index:
  backends:
    exa:
      enabled: true
      api_key: ${EXA_API_KEY}
      scopes: [web, github, research]
    valyu:
      enabled: true
      api_key: ${VALYU_API_KEY}
      required_for: [research, deep_research]

See: skill://valyu for complete recursive research documentation.

Index Types

Index Content Tool
codebase Source code (agent, core, plugins) ck
documentation Docs, ADRs, architecture leann
ideas Ideas, brainstorming, vision docs leann
sessions Session notes, discussions leann
tasks bd issues and TODOs bd
skills POC skills from ~/lev/workshop/poc/lookup lookup CLI
memory AutoMem stored memories (decisions, context) memory_search
web External web search via Exa neural search exa-plus skill

Memory Index (AutoMem Integration)

The memory index connects to Clawdbot's AutoMem extension:

# Search memories
lev get "authentication decision" --indexes memory

# Combined search (code + memory)
lev get "why did we choose JWT" --indexes codebase,documentation,memory

Direct AutoMem tools (available via clawdbot skill):

  • memory_search - Semantic search across stored memories
  • memory_get - Retrieve specific memory by ID
  • memory_store - Store new memory
  • memory_forget - Delete memory

Bash fallback (when MCP unavailable):

# Direct AutoMem API query
curl -s "http://localhost:8001/search?q=authentication&limit=5" | jq '.results'

# Store a memory
curl -X POST "http://localhost:8001/store" \
  -H "Content-Type: application/json" \
  -d '{"content": "We chose JWT for stateless auth", "tags": ["decision", "auth"]}'

Session history search (conversation grep):

# Find in recent sessions
find ~/.clawdbot/agents/main/sessions -name "*.jsonl" -mtime -1 \
  -exec grep -l "keyword" {} \;

# Parse session content
jq -r 'select(.type=="message") | .message.content[0].text // empty' \
  < ~/.clawdbot/agents/main/sessions/{ID}.jsonl | grep -i "keyword"

Search Patterns

Finding Code Patterns

lev get "authentication middleware implementation"
lev get "how to handle errors" --indexes codebase

Finding Architectural Decisions

lev get "why hexagonal architecture" --indexes documentation
lev get "ADR memory system" --indexes documentation

Finding Related Work

lev get "what exists for caching" --indexes codebase,documentation,tasks
lev get "previous discussion about X" --indexes sessions

Finding Tasks

lev get "blocked on oauth" --indexes tasks
lev get "TODO refactor" --indexes codebase,tasks

Finding Skills (Decision Frameworks, Patterns)

lev get "decision making" --indexes skills
lev get "how to handle complexity" --indexes skills,documentation
lev get "debugging patterns" --indexes skills,codebase

Understanding Results

Results are fused using RRF (Reciprocal Rank Fusion):

  • Results from multiple sources are merged
  • Duplicate paths are deduplicated
  • Final ranking reflects consensus across indexes

Output includes:

  • path - File or issue location
  • score - RRF fusion score (higher = more relevant)
  • source - Which index matched
  • snippet - Relevant text excerpt

Routing Rules

The manifest auto-routes queries:

  • TODO|FIXME|task → tasks index
  • idea|vision → ideas + _global
  • Default → codebase + documentation + _global

Override with --indexes for explicit control.

Integration with bd

Search creates trackable jobs:

# View search history
bd list --label find-job

# Resume interrupted search
lev get --resume job-a3f8e9

Tips

Narrow Results

# Too many results? Add specificity
lev get "JWT token refresh" --indexes codebase

# Or filter by path patterns
lev get "auth" | grep middleware

Broaden Results

# Missing results? Search more indexes
lev get "caching" --indexes codebase,documentation,ideas,tasks

Domain Expansion Ladder (When Stuck)

If initial search fails, climb the abstraction ladder:

Level 1: File/Function    → "find auth.ts" (exact match)
Level 2: Component        → "authentication module" (scope wider)
Level 3: Topic            → "auth patterns" (pattern match)
Level 4: Similar          → "how NextAuth does this" (prior art)
Level 5: Goals            → "secure user sessions" (problem space)
Level 6: Ideas            → "zero-trust principles" (concepts)

Usage: Start at your certainty level, climb UP when results sparse:

# Level 1 failed? → Expand to Level 2
lev get "bd-daemon" --indexes codebase       # 0 results
lev get "daemon sync module" --indexes codebase,documentation  # 29 results

# Level 2 too specific? → Climb to Level 3
lev get "auto-sync patterns" --indexes codebase,ideas,sessions

# Unfamiliar domain? → Jump to Level 5-6
lev get "what problem does caching solve" --indexes ideas,sessions

Rule: Level 2 (component) is typically the sweet spot for codebase searches.

Combine with Other Tools

# Find then read
lev get "error handling" | head -5  # Get top results
# Then use Read tool on specific files

# Find then grep for specific lines
lev get "auth middleware" --indexes codebase
# Then grep -n "function" in the matched files

Query Expansion Features

Empty Query: Random Skill Discovery

When searching the skills index without a query, get 5 random skills for exploration:

lev get "" --indexes skills
# Returns 5 random skills from the POC catalog

Use case: Discover new decision frameworks and patterns you didn't know existed.

Related Query Suggestions

After search results, lev-find suggests 2-3 related queries to explore:

lev get "authentication patterns" --indexes skills

# Results shown...

💡 Related searches you might find useful:
  - "authorization access control"
  - "session management security"
  - "identity verification methods"

Skill Categories for Exploration

Browse skills by category to understand the landscape:

# Categories available:
- dev         (Software engineering, patterns, architecture)
- strategy    (Business strategy, competitive analysis)
- cognitive   (Mental models, biases, heuristics)
- decision    (Decision-making frameworks)
- systems     (Systems thinking, complexity)
- design      (UI/UX, product design)

# Example: Explore all decision-making skills
lev get "decision" --indexes skills
# Then use category filter if integrated with lookup CLI

Related Search Tools

lev-find is the unified orchestrator. For specialized searches, use:

Tool Specialty Use When
lev-find (this) Unified local + external search Default choice, cross-domain discovery
lev-research Multi-perspective orchestration Architecture analysis, research workflows
deep-research Tavily multi-query synthesis Complex research, iterative refinement
valyu Recursive turn-based research Confidence-driven, automatic query expansion
brave-search Quick web search Fast lookups, documentation
tavily-search AI-optimized single search Clean snippets, current info
exa-plus Neural web search GitHub, LinkedIn, research papers
grok-research X/Twitter, real-time Social sentiment, trending topics
firecrawl Web scraping Content extraction, site mapping
qmd Local markdown/JSONL search Session history, conversation search
agent-browser Interactive web automation Form filling, testing, screenshots

Decision tree:

Need to search? → Start here:
├─ Local (code/docs/sessions) → lev get
├─ External web (quick) → brave-search or tavily-search
├─ External web (deep) → deep-research or valyu
├─ Social/real-time → grok-research
├─ Structured scraping → firecrawl
├─ Interactive browser → agent-browser
└─ Multi-perspective → lev-research

See also:

  • skill://lev-research - Main research orchestrator
  • skill://deep-research - Tavily deep research
  • skill://valyu - Recursive research

Quick Reference

Command Purpose
lev get "query" Search all default indexes
lev get "query" --indexes X,Y Search specific indexes
lev get "" --indexes skills Random skill discovery (5 samples)
lev get --resume ID Resume interrupted search
bd list --label find-job Search history

Integration with lookup CLI

The skills index is powered by ~/lev/workshop/poc/lookup CLI:

# Direct CLI usage (alternative to lev get)
cd ~/lev/workshop/poc/lookup
./cli.js find "decision making"
./cli.js list --tag=decision
./cli.js load <skill-id>

See lookup CLI documentation for advanced features like hybrid search, tag filtering, and skill loading.

Relates

Master Router

  • Lev Master Router (lev/SKILL.md) - Routes all lev-* skills Parent skill that dispatches to this skill based on keywords/context

Technique Map

  • Role definition - Clarifies operating scope and prevents ambiguous execution.
  • Context enrichment - Captures required inputs before actions.
  • Output structuring - Standardizes deliverables for consistent reuse.
  • Step-by-step workflow - Reduces errors by making execution order explicit.
  • Edge-case handling - Documents safe fallbacks when assumptions fail.

Technique Notes

These techniques improve reliability by making intent, inputs, outputs, and fallback paths explicit. Keep this section concise and additive so existing domain guidance remains primary.

Prompt Architect Overlay

Role Definition

You are the prompt-architect-enhanced specialist for lev-find, responsible for deterministic execution of this skill's guidance while preserving existing workflow and constraints.

Input Contract

  • Required: clear user intent and relevant context for this skill.
  • Preferred: repository/project constraints, existing artifacts, and success criteria.
  • If context is missing, ask focused questions before proceeding.

Output Contract

  • Provide structured, actionable outputs aligned to this skill's existing format.
  • Include assumptions and next steps when appropriate.
  • Preserve compatibility with existing sections and related skills.

Edge Cases & Fallbacks

  • If prerequisites are missing, provide a minimal safe path and request missing inputs.
  • If scope is ambiguous, narrow to the highest-confidence sub-task.
  • If a requested action conflicts with existing constraints, explain and offer compliant alternatives.
Weekly Installs
4
Repository
lev-os/agents
First Seen
14 days ago
Security Audits
Gen Agent Trust HubWarn
SocketWarn
SnykWarn
Installed on
opencode4
antigravity4
claude-code4
github-copilot4
codex4
kimi-cli4