paoloanzn/free-code

The README explicitly states that 'security-prompt guardrails' and 'hardcoded refusal patterns' have been stripped from the upstream Claude Code binary. This is a deliberate modification to remove AI safety restrictions, which is a significant red flag regardless of whether the underlying code is functional.

The install script uses a 'curl | bash' one-liner pattern (curl -fsSL https://raw.githubusercontent.com/.../install.sh | bash), which executes arbitrary remote code without verification. Combined with the project's stated goal of bypassing safety systems, this pattern is particularly risky.

The project claims to be built from source code 'publicly exposed through a source map in Anthropic's npm distribution,' which strongly implies the code is derived from proprietary, non-publicly-licensed software obtained without authorization. The license section effectively acknowledges this, stating the source 'is the property of Anthropic.' Distributing modified proprietary code stripped of safety features is a serious concern.

The repository includes an IPFS mirror specifically noted as a censorship-resistance mechanism ('If this repo gets taken down, the code lives on'), suggesting awareness that the repository may violate terms of service or legal restrictions and is designed to persist despite takedown attempts.

The README contains language that may be intended to normalize bypassing AI safety systems ('The model's own safety training still applies -- this just removes the extra layer of prompt-level restrictions'), which could be designed to reduce analyst or user skepticism about the safety implications.