The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/quartz-bootstrap.sh downloads the Quartz repository from https://github.com/jackyzha0/quartz.git using git clone. While Quartz is a popular tool for documentation, it is hosted in an individual user's repository rather than a verified organization.
[REMOTE_CODE_EXECUTION]: Following the download, scripts/quartz-bootstrap.sh executes npm install and npx quartz within the external repository's directory. This results in the execution of unverified third-party code on the user's machine.
[COMMAND_EXECUTION]: The skill performs several command-line operations: it uses git to analyze the project history and check for file changes. It also provides an optional feature to install a post-commit Git hook, which automatically triggers a Node.js linting script (scripts/graph-lint.mjs) whenever a commit occurs. This serves as a persistence mechanism to maintain the knowledge graph's state.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests and summarizes untrusted content from the target codebase, such as source code, manifest files, and documentation. Maliciously crafted comments or metadata within a scanned project could attempt to influence the agent's summary generation or the results of subsequent queries against the knowledge base. The analysis of these files lack explicit sanitization or boundary markers to isolate analyzed code from agent instructions.

understand-codebase