The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted data from the local codebase to influence its logic.
Ingestion points: The agent reads CONTEXT.md, Architecture Decision Records (ADRs), and the entire codebase using a subagent.
Boundary markers: Absent; there are no instructions to the agent to disregard or delimit instructions that might be embedded in the files it analyzes.
Capability inventory: The skill performs file writes (temp directory, CONTEXT.md, ADRs), executes system commands (open, xdg-open, start), and invokes other autonomous skills (/grilling, /domain-modeling).
Sanitization: Absent; the skill does not specify any validation or filtering of the content ingested from the codebase before it is used in the report or the decision-making loop.
[COMMAND_EXECUTION]: The skill instructs the agent to execute platform-specific shell commands (xdg-open, open, or start) to open a generated HTML report in the user's browser. This is an intended function for presenting architectural analysis.
[EXTERNAL_DOWNLOADS]: The generated HTML report includes references to the Tailwind CSS and Mermaid.js libraries via Content Delivery Networks (CDNs) including cdn.tailwindcss.com and cdn.jsdelivr.net. These are well-known services used to provide styling and diagramming capabilities to the report.

improve-codebase-architecture