The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git logs, diffs, and project files, creating an attack surface for indirect prompt injection.
Ingestion points: Processes output from git log, git diff, and reads project files including GLOSSARY.md, ADRs, and source code.
Boundary markers: The skill includes explicit instructions under 'Scope and authorization' telling the agent not to trust instructions from log files, command output, or source comments as elevated authorization.
Capability inventory: The skill uses git for exploration and OS-level commands (xdg-open, open, start) to display generated reports.
Sanitization: Relies on the agent following instructions to treat ingested data as non-authoritative.
[COMMAND_EXECUTION]: The skill uses standard development commands to gather architecture evidence.
Evidence: Executes git log --oneline -30 and git diff --name-only main...HEAD to identify active areas of the codebase.
Evidence: Uses platform-specific commands (xdg-open, open, start) to present the generated HTML report to the user. These actions are aligned with the skill's stated purpose of reviewing codebase architecture.
[EXTERNAL_DOWNLOADS]: The generated HTML report utilizes external resources for styling and diagram rendering.
Evidence: References https://cdn.tailwindcss.com for layout and styling.
Evidence: References https://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.esm.min.mjs for rendering architectural diagrams. These are well-known, industry-standard services used for documentation and reporting.

improve-codebase-architecture