The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from task.md, questions.md, and research.md. While these ingestion points represent an attack surface for indirect instructions, the skill implements a mandatory human-in-the-loop workflow. The agent must present design questions and options for user approval (Step 3) before writing the design.md file, which mitigates the risk of the agent acting autonomously on injected instructions.
[DATA_EXPOSURE]: The skill reads and writes files within the local project environment. There are no network operations, hardcoded credentials, or attempts to access sensitive system files like SSH keys or environment variables.
[REMOTE_CODE_EXECUTION]: The skill mentions spawning codebase-pattern-finder or codebase-analyzer agents for targeted exploration. These are internal tool calls to other skills within the agent environment and do not involve downloading or executing code from external, untrusted sources.
[COMMAND_EXECUTION]: No shell commands or system-level execution patterns were detected. The skill is focused entirely on reading, analyzing, and writing markdown documentation.

qrspi-design