The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes external CLI tools, specifically gemini and codex, to perform codebase analysis and architectural design tasks. The Architect subagent is instructed to use codex exec with a --full-auto flag and a --sandbox read-only constraint to process architectural questions.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests and processes untrusted data from the local codebase.
Ingestion points: Untrusted data enters the agent context in SKILL.md when the gemini CLI tool analyzes the codebase and saves the findings to .claude/docs/research/{feature}-codebase.md.
Boundary markers: The skill does not implement explicit boundary markers or delimiters to separate analyzed codebase content from agent instructions in subsequent phases.
Capability inventory: The skill has the capability to write files (via SKILL.md instructions for Researcher and Architect) and execute CLI commands (gemini, codex).
Sanitization: There is no evidence of sanitization or filtering of the codebase content before it is used to prompt the Researcher and Architect agents in Phase 2.

startproject