Skills
Navigate
skills/danielrosehill/claude-knowledge-documentation-plugin/new-workspace/Gen Agent Trust Hub

new-workspace

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill interpolates user-supplied input from $ARGUMENTS into bash commands without sanitization, creating a surface for indirect prompt injection. \n
  • Ingestion points: User-provided strings in $ARGUMENTS and identity facts from ~/.claude/CLAUDE.md. \n
  • Boundary markers: None present; values are directly inserted into command strings. \n
  • Capability inventory: Bash execution via mkdir, cp, git, and gh. \n
  • Sanitization: Absent; there are no instructions to escape or validate user inputs before execution. \n- [COMMAND_EXECUTION]: The procedure executes multiple shell commands to scaffold directories, copy templates, and manage Git repositories. This execution surface is tied to the skill's primary purpose but remains a point of concern if arguments are manipulated. \n- [DATA_EXFILTRATION]: Local identity facts are harvested from ~/.claude/CLAUDE.md and pushed to a public GitHub repository by default via gh repo create. While functional, this behavior risk exposing personal information if the user does not specify the --private or --local-only flags.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 06:23 AM
Security Audit — agent-trust-hub — new-workspace