/meaning-query — Semantic Query Engine

Natural language search through the .meaning/ semantic index. Ask questions about the codebase and get structured answers using tags, relationships, intents, and concepts.

Purpose

This skill provides instant semantic search without needing to grep, read files, or guess locations. The meaning index already contains structured metadata—this skill lets you query it conversationally.

When to use:

• •"Where is X implemented?"
• •"What tests file Y?"
• •"Show me all configuration files"
• •"What needs review?"
• •"Files that do authentication"
• •"What changed recently?"

Key benefit: Zero-latency semantic search (no LLM calls, pure structured queries).

Usage

# User asks Claude: "Where is parsing done?"

# Claude runs:
python -m meaning query "where is parsing done?"

The query engine automatically:

1. •Parses natural language intent
2. •Maps to structured queries (tags, relationships, concepts)
3. •Returns relevant files with context

Supported Query Types

1. Status Queries

Ask about file health and maintenance needs.

Examples:

• •"what needs review?"
• •"what is stale?"
• •"show me files that need attention"

Returns: Files filtered by needs_review or is_stale() flags.

2. Tag Queries

Find files by their semantic tags.

Examples:

• •"show me all test files"
• •"find config files"
• •"list API files"
• •"show me documentation"

Returns: Files tagged with matching vocabulary (test, config, api, doc, etc).

3. Relationship Queries

Traverse the semantic graph.

Examples:

• •"what tests the core module?"
• •"what documents the API?"
• •"what imports src/meaning/meaning_core.py?"
• •"show me files that implement the spec"

Returns: Files connected via typed relationships (tests, documents, imports, implements, configures, calls).

4. Intent Queries

Search by what files do (semantic purpose).

Examples:

• •"files that do parsing"
• •"files about authentication"
• •"validation files"
• •"files that handle errors"

Returns: Files whose intent strings match keywords.

5. Temporal Queries

Find recently modified files.

Examples:

• •"what changed recently?"
• •"show me latest files"
• •"recently updated files"

Returns: Files sorted by last_verified timestamp (most recent first).

6. Concept Queries

Explore semantic groupings.

Examples:

• •"show me the core library"
• •"files in the testing concept"
• •"what's in project documentation?"

Returns: All files in a named concept with its description.

Output Format

Query results include:

• •File paths with status indicators (⚠️ for needs_review)
• •Intents (truncated to 80 chars)
• •Tags (first 5 shown)
• •Relationships (summarized by type)
• •Explanation of what was matched
• •Query type detected (status, tag, relationship, intent, temporal, concept)

Example Session

User: Where is parsing implemented?

Claude: Let me search the semantic index for parsing-related files.

[Runs: python -m meaning query "files that do parsing"]

🔍 Query Results: Files tagged with: parsing
   Type: tag

  Found 2 files (showing 2):

  1. src/meaning/meaning_core.py
      "Core Python library implementing YAML parsing, validation..."
      Tags: core, module, api, parsing, validation
      Relationships: implements(1)

  2. src/meaning/meaning_inference.py
      "Inference engine that automatically generates semantic metadata..."
      Tags: core, module, api, parsing
      Relationships: imports(1), implements(1)

Parsing is primarily implemented in `src/meaning/meaning_core.py`, with additional
parsing logic in `src/meaning/meaning_inference.py`. Both files are part of the
core library concept.

Implementation

import subprocess
import json
from pathlib import Path

def run_query(query: str) -> dict:
    """Run a semantic query and parse results."""
    result = subprocess.run(
        ["python", "-m", "meaning", "query", query],
        capture_output=True,
        text=True,
        cwd=Path.cwd()
    )

    if result.returncode != 0:
        return {"error": result.stderr}

    return {"output": result.stdout}

# Example usage in skill
query = "what tests the core?"
result = run_query(query)
print(result["output"])

Query Patterns Reference

Tips for Claude

1. •

   Always use exact file paths in relationship queries

   • •Good: "what tests src/meaning/meaning_core.py"
   • •Bad: "what tests the core" (works but less precise)
2. •

   Use natural language—don't over-think it

   • •The query engine handles variations automatically
   • •"test files", "testing files", "tests" all work
3. •

   Check query type in output

   • •Tells you which matching strategy was used
   • •Helps refine follow-up queries
4. •

   Combine with file reads

   • •Query finds candidates
   • •Read specific files for details
   • •Much faster than reading everything
5. •

   Fallback gracefully

   • •If query returns nothing, try broader terms
   • •"files about X" works when tag queries don't

When NOT to Use

• •Needle queries: Looking for a specific class/function name → use Grep
• •Full-text search: Searching code content → use Grep
• •File existence: Checking if a file exists → use Glob or Read
• •Exploring raw files: Reading unindexed files → use Read directly

Use this skill for semantic understanding, not code searching.

Integration with Other Skills

# Typical workflow:

1. /meaning-query "what tests the API?"
   → Find test files semantically

2. Read the test files
   → Understand test coverage

3. /meaning-update
   → Sync any new files after changes

4. /meaning-validate
   → Verify index health

Error Handling

# No .meaning/ directory found
❌ No .meaning/ directory found in /path/to/project
💡 Initialize with: python -m meaning init

# No results found
🔍 Query Results: No files found matching query: 'nonexistent'
   Type: no_match

  No files found.

Performance

• •Query latency: < 50ms (pure Python, no LLM calls)
• •Scales to: ~5000 files efficiently
• •Memory: Loads full index into memory (~1MB per 1000 files)

Philosophy

Semantic search beats full-text search when:
- You care about purpose, not syntax
- You want structured results, not string matches
- You need relationships, not just content
- You value speed over exhaustiveness

The meaning index is intentionally curated—it's not exhaustive, it's semantic. Use Grep for exhaustive searches, use /meaning-query for understanding.

Last updated: 2026-01-21