Search Engine搜索引擎

Setup

On first use, read setup.md and establish activation behavior, system scope, and data constraints before proposing implementation steps.

When to Use

User needs to create, redesign, or scale a search engine for applications, documentation, products, or internal knowledge bases. Agent handles architecture planning, indexing strategy, retrieval design, relevance controls, evaluation loops, and rollout safety.

Architecture

Memory lives in ~/search-engine/. See memory-template.md for baseline structure and status values.

CODEBLOCK0

Quick Reference

Use the smallest relevant file for the task.

Topic	File
Setup and activation behavior	INLINECODE3
Memory template and status model

memory-template.md |
| Architecture options and component choices | architecture-blueprint.md |
| Retrieval and ranking strategy patterns | retrieval-patterns.md |
| Quality measurement and evaluation loops | evaluation-metrics.md |
| Delivery and rollout gates | implementation-checklist.md |

Data Storage

Local notes stay in ~/search-engine/:

• - requirements and relevance objectives
• data source assumptions and indexing decisions
• experiment outcomes and deployment safeguards

Core Rules

1. Start with a Retrieval Contract, Not with Tools

Before selecting engines, define the contract:

• - query types to support (keyword, phrase, semantic, hybrid)
• response format, latency budget, and freshness target
• error tolerance and fallback behavior

A search engine without a contract becomes an untestable collection of features.

2. Design Ingestion and Indexing as a Deterministic Pipeline

Every document should pass explicit stages:

• - ingestion source validation and deduplication
• normalization and field extraction
• chunking policy with stable identifiers
• indexing with repeatable transforms

Deterministic pipelines reduce drift between environments and simplify debugging.

3. Separate Recall Layers from Precision Layers

Treat retrieval as a staged system:

• - broad candidate retrieval first (lexical, vector, or hybrid)
• reranking and business rules second
• formatting and explanation last

Mixing all concerns in one step hides failures and makes tuning unpredictable.

4. Define Relevance Features as Versioned Policy

Relevance changes must be tracked as policy versions:

• - feature weights and boosts
• typo tolerance and synonym policy
• filtering, faceting, and tie-break rules

Never ship silent relevance changes without versioned notes and measured deltas.

5. Evaluate Offline Before Production Writes

For each relevance or indexing change:

• - run benchmark queries with labeled expectations
• measure hit quality, ordering quality, and coverage
• compare against current baseline and note regressions

If evaluation evidence is weak, keep the current configuration and iterate.

6. Build Idempotent Index Operations and Safe Rollback

Index updates must be replay-safe:

• - stable document ids and version checks
• resumable batch jobs with checkpoints
• alias-based or dual-index rollback plan

Without idempotency and rollback, incident recovery becomes guesswork.

7. Match Complexity to Workload Reality

Use the minimum architecture that meets requirements:

• - avoid distributed complexity for small datasets
• avoid simplistic models for multilingual or high-noise corpora
• revisit design as scale and usage patterns change

Over-engineering and under-engineering both create expensive rework.

Common Traps

• - Starting with vendor selection before defining retrieval requirements -> architecture lock-in with unclear success criteria
• Indexing raw data without field-level normalization -> poor filters, weak facets, and noisy matching
• Tuning relevance on one happy-path query set -> brittle results in real user traffic
• Applying business boosts without guardrails -> top results become commercially biased and less useful
• Shipping retrieval changes without offline baseline comparison -> regressions discovered only by users
• Running full reindex jobs without resumability -> long outages and partial data corruption
• Ignoring multilingual tokenization differences -> severe precision drop for non-English users

Security & Privacy

Data that leaves your machine:

• - none by default in this instruction set
• only user-approved integration traffic when the user explicitly connects external services

Data that stays local:

• - planning notes and experiment logs under INLINECODE10
• constraints, relevance decisions, and rollback records

This skill does NOT:

• - collect unrelated files or credentials
• require hidden network calls
• bypass user-confirmed environment boundaries

Related Skills

Install with clawhub install <slug> if user confirms:

• - api - Define stable APIs for indexing, querying, and retrieval orchestration
• INLINECODE13 - Implement production indexing and query execution on Elasticsearch
• INLINECODE14 - Ship lightweight retrieval stacks with fast iteration cycles
• INLINECODE15 - Structure implementation workstreams and technical decision logs
• INLINECODE16 - Improve delivery quality with testable architecture and rollout discipline

Feedback

• - If useful: INLINECODE17
• Stay updated: INLINECODE18