The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains no instructions to bypass safety filters or override system constraints. It uses structured logic and hard constraints (e.g., the 'HARD-GATE' block) to ensure the agent follows a specific design-first workflow, which acts as a safeguard against premature or unintended code execution.
[DATA_EXFILTRATION]: While the skill involves reading project files and git history to establish context, it does not include any commands to exfiltrate data to external servers or untrusted domains. Operations are limited to reading local context and writing design documentation to a specific local path.
[COMMAND_EXECUTION]: The skill instructs the agent to commit design documents to git. This is a standard development task and is handled within the context of project documentation rather than as an arbitrary or unsafe shell command.
[NO_CODE]: The skill is composed entirely of markdown instructions and a Graphviz (DOT) diagram. It does not include scripts, binaries, or external dependencies that could introduce executable vulnerabilities.

brainstorming