The Agent Skills Directory

[SAFE]: The skill is a pure knowledge-based extension providing a development workflow. It contains no executable scripts, binaries, or active code components.
[COMMAND_EXECUTION]: Several reference files, such as references/drift-detection.md and references/tdd-from-intent.md, include example shell commands and bash scripts for auditing codebases and running tests. However, the tank.json manifest explicitly disables subprocess execution (subprocess: false), ensuring the agent cannot execute these commands autonomously. They serve strictly as documentation for the user.
[DATA_EXFILTRATION]: While the skill requests read access to the filesystem (**/*) to perform its intended audit and sync functions, the tank.json manifest denies all outbound network access, preventing the exfiltration of any data retrieved from the project.
[PROMPT_INJECTION]: The instructions focus on structured development processes and do not contain any attempts to bypass safety filters or override system instructions.

@tank/idd