project-spawn

Pass

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bash script (spawn_session.sh) that automates tmux sessions. It creates a session, changes to a project directory, and sends keystrokes to start an AI agent (claude) and provide initial context. This execution is scoped to the skill's stated purpose of session management.
  • [PROMPT_INJECTION]: The skill's handoff mechanism is a surface for indirect prompt injection. It extracts information from the current conversation and writes it to a PROJECT_HANDOFF.md file for use by a subsequent session.
  • Ingestion points: Up to 10 previous messages from the current conversation context.
  • Boundary markers: Absent. The handoff file template does not use specific delimiters or instructions to treat the extracted content as untrusted input.
  • Capability inventory: The agent in the new session has full shell and file system access within the project directory.
  • Sanitization: Absent. The skill relies on the LLM's instructions to manually filter for 'relevant' information during extraction.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 1, 2026, 09:52 AM