spec-workflow-init

Pass

Audited by Gen Agent Trust Hub on Apr 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes various shell commands to perform project environment discovery.\n
  • Evidence: SKILL.md contains commands such as git rev-parse, ls -la, find, and cat package.json used to identify package managers, CI/CD services, and existing project configurations.\n- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted project metadata into its generated instructions.\n
  • Ingestion points: In SKILL.md, the tool reads contents from local project manifest files like package.json, go.mod, and pyproject.toml.\n
  • Boundary markers: Absent. There are no delimiters or specific instructions provided to the agent to treat the interpolated data as untrusted or to ignore embedded commands.\n
  • Capability inventory: The skill has permissions to write files (Write), create directories (mkdir), and modify project convention files (Edit for AGENTS.md or CLAUDE.md).\n
  • Sanitization: Absent. Data extracted from the project environment is directly inserted into template placeholders without validation or escaping.\n- [SAFE]: The skill performs dynamic generation of documentation and configuration files from predefined local templates.\n
  • Evidence: SKILL.md describes reading templates from the references/ directory and replacing placeholders with user-provided or environment-detected values to create files such as issue-to-pr-workflow.md and agent definitions in .claude/agents/ or .codex/agents/.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 4, 2026, 04:52 PM