The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from the user's codebase (source code, ADRs, glossary files) to perform architectural analysis. Ingestion points: Codebase files, UBIQUITOUS-LANGUAGE.md, and repository ADRs are read into context. Capability inventory: Bash shell access, subagent spawning, and local file operations. Sanitization: The skill explicitly requires escaping all codebase-derived text before embedding it in the generated HTML report to prevent injection.
[COMMAND_EXECUTION]: The skill uses local shell commands for developer workflow integration. It uses git commands to retrieve repository metadata. It utilizes mktemp to create secure temporary files and invokes system-specific commands like open or xdg-open to display generated reports. These operations are restricted to the local environment.
[SAFE]: The skill follows security best practices by avoiding external CDNs or remote scripts in its generated HTML reports, reducing the risk of data exfiltration or supply chain attacks. The use of dynamic context injection for repository metadata (git status, log, branch) is benign and standard for this tool's purpose.

improve