The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by processing untrusted content from the workspace without adequate safeguards.
Ingestion points: The explorer and explainer subagents (referenced in references/explorer-prompt.md and references/explainer-prompt.md) use Glob, Grep, and Read tools to gather data from the codebase.
Boundary markers: The prompt templates for these agents do not utilize boundary markers or explicit instructions to treat file content as data only, potentially allowing natural language instructions in analyzed files to influence agent behavior.
Capability inventory: The skill utilizes general-purpose subagents with the ability to search and read throughout the user's workspace.
Sanitization: No sanitization, validation, or escaping is performed on the ingested code before it is passed to the models for analysis.

how