git:20261110.0e87468 to git:20261110.5be8a6c

34 added, 22 removed. Audit A to A.

What this update changes in the checks

No check changed: every fixed check gives the same answer before and after this update.

A tripped check is a label, not a verdict: read the matched text and the lines below. This week's updates that newly trip a check.

---
name: bigquery-patent-search
description: Fast, cloud-based patent searching across 100 million+ worldwide patents using Google BigQuery - keyword search, CPC classification, patent details retrieval
tools: Bash, Read, Write
model: sonnet
---
# BigQuery Patent Search Skill
- > **Windows:** use `$HOME\patent-creator-venv\Scripts\python.exe` and `$HOME\patent-creator-venv\Scripts\patent-creator.exe` wherever this page says `~/patent-creator-venv/bin/python` or `~/patent-creator-venv/bin/patent-creator`.
+ > **Windows PowerShell:** replace a macOS/Linux Python invocation with `& "$HOME\patent-creator-venv\Scripts\python.exe"` and a CLI invocation with `& "$HOME\patent-creator-venv\Scripts\patent-creator.exe"`. Keep script paths quoted. Python code blocks are Python source, not shell commands.
Fast, cloud-based patent searching across 100 million+ worldwide patents using Google BigQuery.
## When to Use
Invoke this skill when users ask to:
- Search for prior art patents
- Find patents in a specific technology area
- Search by CPC classification code
- Look up patent details by publication number
- Conduct freedom-to-operate searches
- Research patent landscapes
## What This Skill Does
Provides access to Google's public patent dataset:
1. **Keyword Search** across 100M+ patents:
- Full-text search of titles, abstracts, claims
- Filter by country (US, EP, JP, CN, etc.)
- Filter by filing/grant date ranges
- Fast cloud-based queries (< 5 seconds)
2. **CPC Classification Search**:
- Search by CPC code (e.g., "G06F16/", "H04L29/06")
- Browse patent classifications
- Find patents in specific technical domains
3. **Patent Details Retrieval**:
- Get full patent text by publication number
- Access title, abstract, claims, description
- View CPC codes, inventors, assignees
- See filing and grant dates
## Required Setup
This skill requires Google Cloud authentication:
**Prerequisites**:
1. Google Cloud Project (free to create)
- 2. BigQuery API enabled (free for reasonable usage)
+ 2. BigQuery API enabled (queries consume scan quota)
3. Application Default Credentials configured
**Setup Commands**:
```bash
# Install Google Cloud SDK (if not installed)
# Visit: https://cloud.google.com/sdk/docs/install
# Authenticate
gcloud auth application-default login
# Set project (get ID from console.cloud.google.com)
- export GOOGLE_CLOUD_PROJECT=your-project-id
+ "$HOME/patent-creator-venv/bin/patent-creator" config set GOOGLE_CLOUD_PROJECT your-project-id
```
- **Environment Variable**:
- Set in `.env` file: `GOOGLE_CLOUD_PROJECT=your-project-id`
+ Windows PowerShell:
+ ```powershell
+ & "$HOME\patent-creator-venv\Scripts\patent-creator.exe" config set GOOGLE_CLOUD_PROJECT your-project-id
+ ```
+ Restart Claude Code after saving the project. BigQuery is optional; the **Patent Search** skill uses SerpApi by default.
+
## How to Use
When this skill is invoked:
1. **Initialize BigQuery searcher**:
```python
import os
import sys
sys.path.insert(0, os.environ.get('CLAUDE_PLUGIN_ROOT', '.'))
from mcp_server.bigquery_search import BigQueryPatentSearch
searcher = BigQueryPatentSearch()
```
2. **Search by keywords**:
```python
results = searcher.search_by_keywords(
query="blockchain authentication",
limit=20,
country="US", # Optional: filter by country
start_year=2020, # Optional: filter by year
end_year=2024
)
```
3. **Search by CPC code**:
```python
results = searcher.search_by_cpc(
cpc_code="G06F16/", # CPC prefix
limit=20,
country="US"
)
```
4. **Get patent details**:
+ Description retrieval is opt-in and can scan over 1 TiB, exceeding the default
+ 350 GiB cap. Use the free dry-run estimate and obtain approval for any cap
+ increase before retrying. Omit `include_description` for a smaller lookup.
```python
patent = searcher.get_patent_details(
- patent_number="US10123456B2" # Publication number
+ patent_number="US10123456B2", # Publication number
+ include_abstract=True,
+ include_claims=True,
+ include_description=True # Large scan; opt in only when needed
)
```
## BigQuery Dataset
Uses `patents-public-data.patents` on Google BigQuery:
- **100M+ worldwide patents**
- **12M+ US patents** with full text
- Updated weekly
- - Free to query (no billing for reasonable usage)
+ - Each query consumes scan quota. A default keyword search scans hundreds of GiB.
## Search Result Format
- Each result includes:
+ Keyword search returns these keys (unavailable dates are null):
```python
{
- "publication_number": "US10123456B2",
+ "patent_number": "US-10123456-B2",
+ "application_number": "US-201615123456-A",
"title": "Method and system for...",
"abstract": "A system for...",
"filing_date": "2019-01-15",
"grant_date": "2020-06-30",
+ "publication_date": "2020-06-30",
"country": "US",
- "cpc_codes": ["G06F16/245", "H04L29/06"],
- "inventors": ["John Doe", "Jane Smith"],
- "assignee": "Example Corp"
+ "family_id": "12345678",
+ "relevance_score": 4
}
```
- Full patent details also include:
- - `claims`: Full text of all claims
- - `description`: Complete description section
- - `priority_date`: Earliest priority date
- - `family_id`: Patent family ID
+ Detail lookups return the identifiers, title, dates, country and family ID above,
+ plus `cpc_codes` and `ipc_codes`, but no `relevance_score`. They include `claims`
+ by default. `abstract` and `description` require their respective `include_*`
+ flags. Text sections are strings. Inventors, assignees and priority dates are
+ not returned by this implementation. Missing requested text may be empty.
## Presentation Format
Present search results as:
```
PATENT SEARCH RESULTS
====================
Query: "blockchain authentication"
Found: 247 patents (showing top 20)
Date Range: 2020-2024
Country: US
[1] US10123456B2 - System for blockchain-based authentication
Assignee: Example Corp
Filed: 2019-01-15 | Granted: 2020-06-30
CPC: G06F16/245, H04L29/06
Abstract: A system for authenticating users using blockchain
technology with distributed ledger verification...
[2] US10234567B1 - Method of secure authentication using blockchain
...
---
Top 5 Most Relevant:
1. US10123456B2 (95% relevance)
2. US10234567B1 (92% relevance)
...
```
## Advanced Search Techniques
1. **Boolean Operators** in queries:
- "blockchain AND authentication"
- "encryption OR cryptography"
- "(mobile OR wireless) AND security"
- 2. **Phrase Search**:
- - "distributed ledger technology"
- - "public key infrastructure"
+ 2. **Phrase Search** (include the double quotes in the query string):
+ - `query='"distributed ledger technology"'`
+ - `query='"public key infrastructure"'`
3. **CPC Code Hierarchies**:
- "G06F" = Computing
- "G06F16/" = Information retrieval
- "G06F16/245" = Structured query language
## Common CPC Codes
- **G06F**: Computing, calculating, counting
- **H04L**: Digital communication
- **G06Q**: Business methods
- **H04W**: Wireless communication
- **G06N**: Computer systems based on specific models
- **G06T**: Image processing
## Error Handling
If BigQuery is not configured:
1. Check if `google-cloud-bigquery` is installed
2. Verify authentication: `gcloud auth application-default login`
3. Confirm project ID in environment: `GOOGLE_CLOUD_PROJECT`
- 4. Test with: `~/patent-creator-venv/bin/patent-creator check-bigquery`
+ 4. Test with: `"$HOME/patent-creator-venv/bin/patent-creator" check-bigquery`
## Cost Considerations
BigQuery pricing:
- **First 1 TiB/month**: FREE
- **After 1 TiB**: $6.25 per TiB queried (on-demand)
- - **Typical query**: 10-50 MB per search
+ - **Observed default keyword scan**: roughly 325-341 GiB per search, about three searches per 1 TiB allowance. These are historical measurements; use the dry-run estimate for the current query.
+ - Selecting fewer text fields reduces scan size. Country, date and keyword filters do not reliably reduce bytes scanned on this unpartitioned table.
- **Bytes-billed ceiling**: enforced per-query via `PATENT_BIGQUERY_MAX_BYTES_BILLED` (default 350 GiB)
## Tools Available
- **Bash**: To run Python BigQuery searches
- **Read**: To load saved search results
- **Write**: To save patent search results
- **Grep**: To search through saved results