The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill mandates a workflow where an agent creates pressure scenarios using subagents and subsequently ingests their rationalizations to refine documentation.
Ingestion points: The main agent is instructed to process verbatim rationalizations and failure patterns provided by subagents during the 'RED' phase of skill creation (SKILL.md).
Capability inventory: The skill references capabilities including file system modification via TodoWrite and shell command execution using tools like wc, grep, and git (SKILL.md).
Boundary markers: The instructions lack specific guidance on using delimiters or boundary markers to isolate potentially malicious instructions within the subagent-generated content.
Sanitization: There are no provisions for sanitizing or filtering the natural language feedback received from subagents before it is processed by the primary agent.

Creating Skills