paper-spine-ui

Warn

Audited by Gen Agent Trust Hub on Oct 30, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The configuration wizard scripts/intake_wizard.py uses os.system to execute chcp for character set configuration and cls or clear to manage the terminal display.
  • [COMMAND_EXECUTION]: The PowerShell launcher scripts/launch_paperspine_ui.ps1 uses Start-Process to spawn a new PowerShell instance to run the wizard.
  • [COMMAND_EXECUTION]: The Bash launcher scripts/launch_paperspine_ui.sh uses osascript (macOS) and various terminal emulators like gnome-terminal, konsole, and xterm (Linux) to execute shell commands in a new window.
  • [PRIVILEGE_ESCALATION]: The instructions in SKILL.md direct the agent to request sandbox_permissions: require_escalated to ensure the UI window appears reliably in sandboxed environments.
  • [PRIVILEGE_ESCALATION]: The PowerShell wrapper scripts/launch_paperspine_ui.ps1 and instructions in SKILL.md utilize the -ExecutionPolicy Bypass flag to execute local scripts, bypassing default Windows script execution restrictions.
  • [DYNAMIC_EXECUTION]: The Bash launcher scripts/launch_paperspine_ui.sh dynamically constructs command strings that are executed in new process contexts via terminal utilities or osascript.
  • [INDIRECT_PROMPT_INJECTION]: The script scripts/intake_wizard.py automatically reads Markdown and text files from the user's workspace to infer configuration values like 'motivation' and 'official URLs'. This ingestion of untrusted data represents an attack surface for indirect prompt injection.
  • Ingestion points: scripts/intake_wizard.py reads *.md and *.txt files via read_small_text_files; scripts/humanize_check.py reads all *.tex files in the final_paper directory.
  • Boundary markers: None; raw file contents are processed without delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill can write configuration and report files (paper_spine_config.json, paper_spine_config.md, humanize_report.md) and execute shell commands via its TUI wrappers.
  • Sanitization: No sanitization or validation of the ingested content is performed before it is processed for inference or presented to the user.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 30, 2026, 02:29 AM