The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external content from the project's codebase, including commit history and source files, to generate reports. This creates a surface where malicious instructions embedded in the codebase could attempt to influence the agent's behavior.
Ingestion points: Reads git log output, source code files, GLOSSARY.md, and files within docs/adr/ (SKILL.md).
Boundary markers: Absent. The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested codebase data.
Capability inventory: Spawns sub-agents for exploration, writes HTML files to system temporary directories, executes system commands to open files (open, start, xdg-open), modifies GLOSSARY.md, and creates new ADR documentation files.
Sanitization: Absent. There is no evidence of filtering or escaping logic for data extracted from the codebase before it is included in prompts or the generated HTML report.
[COMMAND_EXECUTION]: The skill executes shell commands to inspect the repository history and to open the generated report in the user's default browser.
Evidence: Executes git log --oneline to find codebase hot spots and uses platform-specific commands (xdg-open, open, start) to display the HTML report (SKILL.md).
[EXTERNAL_DOWNLOADS]: The generated HTML report incorporates external resources from well-known content delivery networks.
Evidence: The HTML scaffold in HTML-REPORT.md fetches the Tailwind CSS library from cdn.tailwindcss.com and the Mermaid diagramming library from cdn.jsdelivr.net.
[DYNAMIC_EXECUTION]: The generated report configures the Mermaid library with a relaxed security posture, which could be exploited if malicious content is injected into the diagram definitions.
Evidence: The initialization script in HTML-REPORT.md sets securityLevel: "loose". This allows HTML tags within Mermaid diagrams, potentially enabling cross-site scripting (XSS) if the AI includes unsanitized strings from the codebase (such as malicious file names or comments) in the diagram output.

improve-codebase-architecture