The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands in scripts/create-design-doc.md and SKILL.md to retrieve project metadata. These include date for timestamps and git commands (git config, git branch, git log, git rev-parse) to identify the author and current repository state. These are benign utility operations used to populate design templates and do not present a security risk.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its project context discovery phase. In SKILL.md and references/phase-workflow.md, the agent is instructed to scan the codebase using Grep, Glob, and Read to determine the project tier. Maliciously crafted content within the analyzed repository (e.g., instructions hidden in a README.md or source comments) could potentially influence the agent's behavior during the brainstorming process. This is documented as a vulnerability surface inherent to codebase-aware agents.
[PROMPT_INJECTION]: The skill instructions in SKILL.md and references/phase-workflow.md use strong steering language (e.g., 'CRITICAL', 'MANDATORY', 'DIVERGENT MODE') to manage the agent's behavior and sub-agent tasking. These instructions are focused on ensuring a thorough brainstorming process and do not attempt to bypass safety guidelines or override system prompts.

brainstorming