# Sgrep

> Semantic and hybrid code and conversation search for intent-based queries. Use when exploring unfamiliar codebases, finding code by concept instead of exact text, or recalling past agent conversations about similar problems.

- **Type:** Skill
- **Install:** `agentstack add skill-xiaoconstantine-sgrep-sgrep`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [XiaoConstantine](https://agentstack.voostack.com/s/xiaoconstantine)
- **Installs:** 0
- **Category:** [Search](https://agentstack.voostack.com/c/search)
- **Latest version:** 0.1.0
- **License:** Apache-2.0
- **Upstream author:** [XiaoConstantine](https://github.com/XiaoConstantine)
- **Source:** https://github.com/XiaoConstantine/sgrep/tree/main/plugins/sgrep/skills/sgrep
- **Website:** http://xiaocui.me/sgrep/

## Install

```sh
agentstack add skill-xiaoconstantine-sgrep-sgrep
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# sgrep - Smart Code & Conversation Search

Use `sgrep` for semantic and hybrid search across **code** and **agent conversations**. It understands intent, not just exact strings.

## When to Use

### Code Search
- Finding code by **concept**: "error handling", "authentication logic", "rate limiting"
- Searching for **specific terms** with semantic context: use `--hybrid`
- Best code-search accuracy after indexing: use `--hybrid --colbert`
- Exploring unfamiliar codebases
- When ripgrep patterns keep missing relevant code

### Conversation Search
- Finding past discussions with **Claude Code**, **Codex CLI**, **Cursor**, **OpenCode**, or **Pi**
- Recalling how you solved a similar problem before
- Building context from previous sessions for new tasks
- Searching across all your coding agent interactions

## Commands

```bash
# First time only
sgrep setup
sgrep setup --with-rerank  # optional, only for --rerank

# Index current directory; builds compact TQ-MSE chunk/file vectors by default
sgrep index .

# Optional ColBERT segment codec override
sgrep index . --colbert-codec tqmse
sgrep index . --colbert-codec int8
sgrep index . --colbert-codec pq6

# Legacy compatibility: also persist full SQL vectors
sgrep index . --sql-vectors

# Watch mode keeps SQL vectors for incremental updates; rerun index to compact
sgrep watch .

# Balanced semantic + lexical code search (default)
sgrep "database connection pooling"
sgrep "how are errors handled"

# Fast semantic-only search
sgrep --profile fast "error handling"

# Best code-search accuracy
sgrep --profile quality "JWT validation"
sgrep --profile quality "authentication middleware"

# With code context
sgrep -c "authentication middleware"

# JSON output
sgrep --json "rate limiting"
```

## Conversation Search

```bash
# Index conversations; refreshes compact TQ-MSE turn vectors
sgrep conv index
sgrep conv index --source codex
sgrep conv index --source claude
sgrep conv index --source opencode
sgrep conv index --source pi
sgrep conv index --watch
sgrep conv index --force

# Search conversations
sgrep conv "authentication flow"
sgrep conv "JWT refresh_token" --hybrid
sgrep conv "database migration" --agent claude --since 7d
sgrep conv "bug fix" --project payment-service --after 2026-01-01 --before 2026-06-01
sgrep conv "exact phrase" --exact
sgrep conv "auth" --json -n 1

# View, export, context, and copy helpers
sgrep conv view 
sgrep conv view  --turn 3 --no-color
sgrep conv export  --format markdown -o conversation.md
sgrep conv export  --format json -o conversation.json
sgrep conv context 
sgrep conv context  --turns 10 --copy
sgrep conv copy  --turn 2 --code-only
sgrep conv status
```

## Semantic vs Hybrid

| Mode | Best For | Example |
|------|----------|---------|
| `--profile fast` | Lowest-latency semantic exploration | "how does auth work" |
| `--profile balanced` (default) | Semantic + exact-term recall | "JWT token validation" |
| `--profile quality` | Highest code-search accuracy | "authentication middleware" |

Use the default balanced profile for most agent searches and `--profile quality` when ranking quality matters more than minimum latency.

## Search Hierarchy

1. **sgrep** → Balanced semantic + lexical discovery
2. **sgrep --profile fast** → Lowest-latency semantic discovery
3. **sgrep --profile quality** → Rerank candidates with precomputed late interaction
4. **ast-grep** → Match structural patterns in those files
5. **ripgrep** → Exact text for specific symbols

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [XiaoConstantine](https://github.com/XiaoConstantine)
- **Source:** [XiaoConstantine/sgrep](https://github.com/XiaoConstantine/sgrep)
- **License:** Apache-2.0
- **Homepage:** http://xiaocui.me/sgrep/

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-xiaoconstantine-sgrep-sgrep
- Seller: https://agentstack.voostack.com/s/xiaoconstantine
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.
