project-documenter

SUSPICIOUS. The core documentation-generation behavior is coherent and there are no strong malware indicators, remote installers, or obvious credential-harvesting proxies. Risk comes from Mode 5 materially expanding scope into executable CI automation and local scripts that use API secrets without fully specifying dependencies, endpoints, or secret handling, making it broader and riskier than a simple documentation skill.

The fragment implements a legitimate AI-assisted documentation CI system and contains no clear malicious payload, credential theft, reverse shell, cryptomining, or covert exfiltration. It does intentionally transmit repository-derived content to Anthropic/OpenAI and can automatically overwrite documentation and push commits with write permissions. Restrict workflow execution for untrusted contributions, minimize token permissions, allowlist writable documentation paths, validate AI output as a constrained patch, pin dependencies and action versions, and require review before automatic pushes.