# Search Docs

> Use when users need source-grounded answers from git-library for API/tool/configuration/migration/troubleshooting questions, or when they ask for library structure/topics overview. Triggers: "search docs", "文档检索", "API 文档", "MCP 配置", "migration guide", "troubleshooting", "最新功能".

- **Type:** Skill
- **Install:** `agentstack add skill-life-is-blue-agent-skills-search-docs`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [life-is-blue](https://agentstack.voostack.com/s/life-is-blue)
- **Installs:** 0
- **Category:** [Search](https://agentstack.voostack.com/c/search)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [life-is-blue](https://github.com/life-is-blue)
- **Source:** https://github.com/life-is-blue/agent-skills/tree/main/skills/search-docs

## Install

```sh
agentstack add skill-life-is-blue-agent-skills-search-docs
```

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

## About

# Search Agentic Knowledge Base

## Goal
Route → retrieve → read source → answer. Minimal calls.

## Config
- Confidence thresholds: high (raw_score ≤ -5), medium (-5  -2)
- Ambiguity: max_gap = 0.20, min_overlap = 0.45
- Budget: default 5 calls, hard cap 8, max 1 cross-library fallback
- Query variants: max 2

## Search Response Schema
Each search result contains:
- `path`: Document path within library (e.g., "guides/setup.md")
- `title`: Document title (nullable)
- `summary`: Short document summary (nullable)
- `excerpt`: Matching text excerpt
- `raw_score`: BM25 raw score (nullable; lower = more relevant, e.g., -8.5)
- `display_score`: Normalized 0-100 score (nullable)
- `highlights`: Array of highlighted matching terms
- `library_id`: Library the result belongs to
- `last_updated`: ISO timestamp of last update (mode=recent only)

Wrapper payload includes:
- `total`: Total matches before pagination
- `limit`, `offset`: Pagination params
- `catalog`: Topic map or full catalog (per catalog_mode)
- `hint`: Human-readable summary of results

## Workflow: Navigate → Search → Probe → Fallback

### Navigate (preferred)
Already know the path? `search-docs read LIB_ID/PATH.md`

### Search (default)
1. Route to one primary library (explicit product/library mention wins).
2. Search inside that library:
```bash
search-docs search "QUERY" --library LIBRARY_ID --limit 8 --catalog-mode none
```
3. Check response confidence. Read top 1-3 docs before answering.

### Probe (ambiguity)
No clear library? Run cross-library probe:
```bash
search-docs search "QUERY" --limit 8
```
If results span multiple libraries with close scores (gap ≤ 0.20):
- Interactive: ask user to clarify.
- Non-interactive: return best + second candidate, mark uncertainty.

### Fallback (once)
Search confidence low? Browse manifest:
```bash
search-docs manifest LIBRARY_ID
```
Navigate topic map → read target doc. No repeated fallback loops.

## Freshness
Queries with "latest/new/recent/最新/刚发布":
```bash
search-docs libraries --fresh-for-query "QUERY"
```
If routing still weak, `search-docs libraries --refresh` then retry once.

## Explore Path
When user asks for structure/topics/coverage (not a concrete answer):
```bash
search-docs libraries
search-docs manifest LIBRARY_ID
```
Deliver: library positioning, topic distribution, recommended starting docs.

## Commands
```bash
search-docs health                                    # connectivity check
search-docs libraries                                 # list all libraries
search-docs libraries --fresh-for-query "QUERY"       # freshness-aware list
search-docs search "Q" --library LIB --limit 8        # targeted search
search-docs search "Q" --limit 8                      # cross-library search
search-docs read LIB/PATH.md                          # read full document
search-docs manifest LIB                              # browse topic map
search-docs recent --days 7                           # recent updates
```

## Anti-Patterns
- Starting with cross-library search for every query — route first.
- Answering from snippets without reading source doc.
- Repeating fallback loops beyond one round.
- Treating static routing hints as stronger than live library metadata.

## Source & license

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

- **Author:** [life-is-blue](https://github.com/life-is-blue)
- **Source:** [life-is-blue/agent-skills](https://github.com/life-is-blue/agent-skills)
- **License:** MIT

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-life-is-blue-agent-skills-search-docs
- Seller: https://agentstack.voostack.com/s/life-is-blue
- 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%.