The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell diagnostic commands such as which and <cmd> --version in Step 5. These commands are used to check for the presence and version of local development tools (e.g., Go, Node.js, Flutter) based on the architectural selection made in previous steps.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing untrusted project documentation through a subagent to drive architectural decisions and environment checks.\n
Ingestion points: The subagent reads project-specific files such as docs/product/business-flow.md, docs/specs/data/, and docs/specs/API.\n
Boundary markers: The subagent prompt template in Step 4.1 does not include explicit delimiters or instructions to ignore potential instructions embedded within the analyzed project files.\n
Capability inventory: The skill has the capability to read local files, write documentation to the docs/standards/ directory, and execute shell commands for software versioning.\n
Sanitization: There is no explicit sanitization or filtering of the subagent's output (specifically the tech stack names) before it is used to construct shell commands for the environment check phase.

architecture