nightshift
Fail
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/nightshift.shscript executes theclaudeandcodextools using the--dangerously-skip-permissionsand--dangerously-bypass-approvals-and-sandboxflags. This bypasses built-in safety mechanisms and allows the agent to execute any shell command without user oversight or approval. - [PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection. Its primary workflow involves reading content from project files such as
docs/product-specs/*.mdanddocs/BUGS.mdand using that content to drive autonomous agent actions. Maliciously crafted documentation or task backlogs could hijack the agent's behavior during unattended sessions to perform unauthorized activities. - [REMOTE_CODE_EXECUTION]: The autonomous loop is designed to write and execute code (such as tests and implementation logic) based on external specifications. Combined with the bypass of security approvals and the lack of human oversight, this provides a dangerous mechanism for arbitrary code execution if the source documentation is compromised.
Recommendations
- AI detected serious security threats
Audit Metadata