Knowledge Lookup

Overview

Query cached external documentation and blog content to get fresh context for development tasks. Knowledge sources are synced on session start and cached for 24 hours.

Core principle: Get authoritative, up-to-date information from official sources.

Trigger: When you need current information about Claude Code, best practices, or configured documentation sources.

When to Use

Invoke this skill when:

User asks about Claude Code features or best practices
You need to reference official documentation
Answering questions about hooks, commands, or Claude Code architecture
Looking for examples from the engineering blog

Available Knowledge Sources

Default sources (synced automatically):

anthropic-engineering - Claude Code Engineering Blog
claude-code-docs-overview - Claude Code Documentation Overview
claude-code-docs-hooks - Claude Code Hooks Reference

Lookup Process

Step 1: List Available Knowledge

Use the Read tool to check what knowledge is available:

Read: ~/.claude/config/knowledge/sources.json

Step 2: Check Cache Freshness

Use bash for SQLite queries (no native equivalent):

sqlite3 ~/.claude/config/knowledge/cache.db \
  "SELECT source_id, fetched_at, expires_at, status FROM knowledge_cache"

Step 3: Read Specific Content

Use the Read tool to read cached content:

Read: ~/.claude/config/knowledge/content/anthropic-engineering.md
Read: ~/.claude/config/knowledge/content/claude-code-docs-overview.md

Step 4: Search Across Sources

Use the Grep tool to search across all cached content:

Grep: pattern="hooks", path="~/.claude/config/knowledge/content/"
Grep: pattern="MCP", path="~/.claude/config/knowledge/content/"

Step 5: Semantic Search (If Available)

When embeddings are initialized, use semantic similarity for better results:

# Check if semantic search is available
from popkit_shared.utils.embedding_store import EmbeddingStore
from popkit_shared.utils.voyage_client import is_available

store = EmbeddingStore()
if is_available() and store.count("knowledge") > 0:
    # Semantic search available
    from popkit_shared.utils.voyage_client import embed_query

    query_embedding = embed_query("how do hooks work")
    results = store.search(
        query_embedding,
        source_type="knowledge",
        top_k=5,
        min_similarity=0.7
    )

    for result in results:
        print(f"[{result.similarity:.2f}] {result.record.source_id}")
        print(f"  {result.record.content[:100]}...")

Hybrid Search Strategy:

Try semantic search first (if embeddings available)
Fall back to Grep for keyword matching
Combine results with weighted scoring:
- Semantic score: 70% weight
- Keyword match: 30% weight

Initialize Embeddings:

python hooks/utils/embedding_init.py --force

Response Format

When providing information from knowledge sources:

## Knowledge Reference

**Source:** [Source Name]
**Fetched:** [Timestamp]
**Relevance:** [Why this is relevant to the query]

### Key Information

[Relevant excerpt or summary from cached knowledge]

---

_From cached knowledge source: [source_id]_
_Last updated: [fetched_at]_

Example Queries

Query: "How do hooks work in Claude Code?"

Use the Read tool to read the hooks documentation:

Read: ~/.claude/config/knowledge/content/claude-code-docs-hooks.md

Extract relevant sections
Provide answer with source attribution

Query: "What are best practices for Claude Code?"

Use the Read tool to read the engineering blog:

Read: ~/.claude/config/knowledge/content/anthropic-engineering.md

Look for best practices content
Summarize with examples

Query: "Is there something about X in the docs?"

Use the Grep tool to search all sources:

Grep: pattern="X", path="~/.claude/config/knowledge/content/"

If found, use Read tool to get the full context
If not found, inform user and suggest they check directly

Handling Stale or Missing Data

If knowledge is stale or missing:

**Note:** The cached knowledge for [source] is [stale/missing].

Last fetched: [date] (expired [time] ago)

Would you like me to:

1. Refresh the knowledge sources? (`/popkit:knowledge refresh`)
2. Provide what I have from my training data?
3. Search the web for current information?

/popkit:knowledge command - Manage knowledge sources
knowledge-sync.py hook - Automatic session start sync
Session start hook - Triggers knowledge sync with TTL check

pop-knowledge-lookup