Does this skill require claudemem to be installed?

Yes, the skill relies on claudemem for indexing and semantic search. It will automatically check the status of your index and prompt you to index the repository if it hasn't been done yet.

How does this skill save tokens during development?

Instead of reading multiple full files which can cost thousands of tokens, it redirects Claude to use semantic search which retrieves only the most relevant ranked chunks, often reducing token usage by up to 90%.

What specifically triggers the Search Interceptor?

It triggers when Claude attempts to read three or more files in the same directory, uses broad glob patterns like **/*.py, or plans sequential reads to understand a concept semantically.

Will it stop me from reading a single specific file?

No. The interceptor decision matrix allows for reading 1-2 specific files or performing exact string matches (grep) without interference. It only intercepts bulk operations where semantic search is objectively more efficient.

Search Interceptor

Name: Search Interceptor
Author: MadAppGang

byMadAppGang

•

177

•

生産性とワークフロー

Optimizes code exploration by intercepting bulk file operations and redirecting Claude to use token-efficient semantic search.

The Search Interceptor skill monitors and redirects Claude's behavior during codebase exploration to prevent inefficient sequential file reads and broad glob patterns. By enforcing a semantic-first protocol using claudemem, it minimizes token consumption and prevents 'tool familiarity bias,' ensuring the AI identifies the most relevant code chunks across the entire repository rather than just the files it manually selects. This skill is essential for maintaining performance and accuracy when working within large, complex projects where manual file traversal is prohibitively expensive.

主な機能

01Broad glob pattern detection and redirection

02Automatic interception of bulk file reads (3+ files)

03Semantic search validation via claudemem

04Token consumption optimization

05Intelligent tool selection decision matrix

06177 GitHub stars

ユースケース

01Preventing token waste when exploring large authentication or API modules

02Breaking AI habits of sequential file reading during initial project onboarding

03Auditing complex codebases where relevant logic is scattered across many directories

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/claude-code search-interceptor

For use in Claude.ai and ChatGPT

Download Skill