The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes system-level commands through the Bash and node tools to execute tests, linting, and formatting (e.g., 'pnpm test'). It also specifies a pattern in 'SKILL.md' for executing local scripts such as '.claude/hooks/routing/routing-guard.cjs' using 'child_process.spawn'.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from user-defined task descriptions and scenario backlogs (ingestion points in 'SKILL.md' and 'schemas/input.schema.json'). These inputs guide agent behavior during the TDD loop without explicit sanitization or markers to 'ignore embedded instructions'. The skill's broad capabilities, including file system access and command execution (inventory in 'SKILL.md' and 'hooks/post-execute.cjs'), present a surface that could be exploited by malicious instructions in test scenarios.

tdd