token-saver-claude-code
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides documentation and guidance for managing token budgets and session efficiency within Claude Code and other coding agents. All external references point to official documentation (e.g., Anthropic's code.claude.com and support.claude.com) or well-known technical platforms (Qiita).
- [COMMAND_EXECUTION]: The skill includes a local utility script (
scripts/audit.mjs) intended to be executed via Node.js. This script is read-only and uses standard Node.js modules without external dependencies. It analyzes local instruction files (like CLAUDE.md) and agent settings to provide optimization recommendations. - [INDIRECT_PROMPT_INJECTION]: The audit script represents a surface for indirect prompt injection as it ingests untrusted project files (
CLAUDE.md,.cursorrules, etc.) to count lines and tokens. However, the script processes this data as text and does not execute the content, and it lacks network capabilities, preventing exfiltration or significant exploitation. The risk is assessed as low/safe due to the read-only nature and lack of dangerous capabilities.
Audit Metadata