tdd-methodology-expert

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Persistence Mechanisms] (MEDIUM): The scripts/setup_hooks.sh script installs a git pre-commit hook and a Claude Code hook. These mechanisms allow the skill's logic to execute automatically during development activities. While central to the intended TDD enforcement use case, git hooks are a standard persistence vector.\n- [Indirect Prompt Injection] (LOW): The assets/hook-templates/user-prompt-submit.sh hook creates a surface for indirect prompt injection by wrapping untrusted user input with methodology reminders and instructions.\n
  • Ingestion points: The user's original prompt is read from stdin into the USER_PROMPT variable in user-prompt-submit.sh.\n
  • Boundary markers: Absent; user input is directly appended at the end of the instruction template without any delimiters or 'ignore embedded instructions' warnings.\n
  • Capability inventory: The hook influences the LLM's response behavior and references an external compliance check script.\n
  • Sanitization: None; user input is not escaped, filtered, or validated before being combined with system-level instructions.\n- [Dynamic Execution] (MEDIUM): The assets/hook-templates/pre-commit.sh script calculates the path to its helper script (SKILL_SCRIPT) at runtime based on absolute and relative paths, then executes it using python3. Additionally, it uses inline Python code to parse JSON output from these scripts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:05 PM