gsd-orchestration

Pass

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it stores user-provided requirements and discussion outcomes in local files (PROJECT.md, CONTEXT.md) which are then used to generate executable task plans. This 'data-as-instructions' pattern is a known vector for indirect injection.\n
  • Ingestion points: .gsd/PROJECT.md, .gsd/phases//CONTEXT.md, and .gsd/phases//PLAN-*.md.\n
  • Boundary markers: Uses XML tags (, , ) to delimit instructions within plan files.\n
  • Capability inventory: Subagent spawning (Task tool), skill loading (Skill tool), and file operations (Read, Write, Edit).\n
  • Sanitization: The skill relies on 'fresh context' protocols to reset the agent's state between tasks, limiting the persistence of malicious instructions.\n- [COMMAND_EXECUTION]: The skill uses Python scripts ('check_state.py', 'validate_plan.py') to perform project maintenance and git-based versioning. These scripts execute shell commands through subprocesses to interact with the local repository, although actions are restricted to standard development utilities.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 28, 2026, 07:53 AM