The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to generate formatted timestamps for its architecture review reports using the date command.
[DATA_EXPOSURE]: The skill requires access to read local design documents and rule files (e.g., in .claude/skills/control-tower/rules/). It also has Write permissions to create new review reports in spec/arch-review/ and automatically update existing specification files in the spec/ directory.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebSearch tool to research architectural patterns, tradeoffs, and comparisons online to ground its evaluation in industry standards.
[PROMPT_INJECTION]: As the skill ingests and processes untrusted design documents provided by the user, it is theoretically subject to indirect prompt injection. The skill lack specific boundary markers or sanitization logic for this ingested content, though its primary function is to analyze the content rather than execute instructions within it.
Ingestion points: Design documents and spec files read from the workspace via Read, Grep, and Glob tools.
Boundary markers: None identified for the processed design document content.
Capability inventory: Includes Bash for timestamps, Write for saving reports and updating project specs, and AskUserQuestion for interactive feedback.
Sanitization: No explicit sanitization or filtering of the ingested design document content is described.

arch-review