The Agent Skills Directory

[PROMPT_INJECTION]: The instructions include coercive language intended to override the agent's reasoning. Specifically, the 'Critical Requirements' section uses phrases like 'MANDATORY: Use Sourcebot MCP 5-7 times' and 'TASK INVALIDATION: Task will be invalidated if you don't use Sourcebot 5-7 times' to force a specific frequency of tool usage.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external code repositories, creating a surface for indirect prompt injection. * Ingestion points: Data returned from the search_code and get_file_source tools in SKILL.md. * Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded instructions in the fetched code. * Capability inventory: The skill uses list_repos, search_code, and get_file_source to interact with external systems. * Sanitization: Absent; the instructions do not require validation or sanitization of the content fetched from GitHub repositories.

sourcebot