The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands for project research and context resolution. It calls a local script node scripts/resolve-feature.js in SKILL.md to identify the target feature. Additionally, the Codex research prompt in references/codex-prompt.md instructs the agent to use tools such as git status, git diff, ls, grep, and cat for repository exploration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from technical specifications and source code to generate architecture designs, representing a surface for indirect prompt injection. Ingestion points: Technical specifications (e.g., 2-tech-spec.md) and codebase files identified during research. Boundary markers: The instructions do not define explicit text delimiters for external content, though they include general safety directives to redact secrets. Capability inventory: The skill possesses broad capabilities including Bash access, file writing, and the ability to invoke other sub-agents and skills. Sanitization: There is no evidence of content validation or escaping for the ingested files before processing by sub-agents.

architecture