The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external artifacts, such as code slices, plans, or PRDs, and includes them in sub-agent prompts. A malicious artifact containing hidden instructions could potentially influence the behavior of the spawned sub-agents. * Ingestion points: Artifact paths, pasted code, and review surfaces provided by the user in .claude/agents/ or via direct input. * Boundary markers: Uses structured headers (e.g., Sub-task:, Source:, Constraints:) to separate instructions from data, and explicitly warns members not to assume other members' conclusions. * Capability inventory: Spawns sub-agents using the platform's Task tool with various capabilities like explore, docs-researcher, and computerUse. * Sanitization: Includes specific instructions for the coordinator to redact secrets before pasting content into member prompts to prevent credential exposure.
[EXTERNAL_DOWNLOADS]: The skill references various markdown files and schemas hosted on the author's GitHub repository (github.com/csark0812/toolbox). These include task-splitting guidelines, output schemas, and domain-specific prompts for different process skills. These references originate from the skill vendor's own infrastructure and are used for task configuration.

council