The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from an external codebase to answer architectural questions, which introduces an attack surface for instructions embedded in code comments or documentation.
Ingestion points: The skill interpolates the user question into subagent prompts and ingests codebase content via ripgrep and file read tools, as defined in references/explorer-prompt.md and references/explainer-prompt.md.
Boundary markers: The prompt templates use markdown headers and blockquotes to delimit the user question and explorer findings.
Capability inventory: The skill utilizes filesystem search and read capabilities (rg, file reads) and can spawn subagents through a model capability router.
Sanitization: There is no evidence of explicit sanitization or content filtering for data retrieved from the codebase before it is passed to the explainer agent.
[EXTERNAL_DOWNLOADS]: The skill metadata in MIRROR.md identifies the source of the skill as the Cursor plugins repository on GitHub.
Evidence: https://github.com/cursor/plugins.git referenced in the mirror metadata.
[COMMAND_EXECUTION]: The exploration instructions direct the agent to utilize command-line search tools to identify symbols and code patterns within the project.
Evidence: The references/explorer-prompt.md file instructs the agent to use rg --files, rg, and file read tools to locate directories and implementations.

cursor-how