The Agent Skills Directory

[SAFE]: The skill follows security best practices for AI agents. It does not contain hardcoded credentials, malicious obfuscation, or unauthorized network activity.
[INDIRECT_PROMPT_INJECTION]: The skill identifies external implementation plans (*.plan.md) as a potential attack vector. It implements specific defensive boundaries, instructing the agent to treat these files as plain data and explicitly disregard common injection phrases like "ignore previous rules" or "skip validation."
[COMMAND_EXECUTION]: The skill manages the execution of local development tools including npm, pnpm, yarn, and bun. It includes mandatory safety gates, such as requiring human review for shell commands and explicitly rejecting destructive operations or remote script execution (e.g., curl | sh).
[PROMPT_INJECTION]: While static analysis tools may flag the mention of "ignore previous instructions," these occurrences are defensive requirements meant to secure the agent against malicious user-provided plan files, rather than attempts to bypass agent safety filters.

tdd-workflow