The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture involves a 'Lead' agent ingesting and synthesizing data from multiple independent 'Teammate' sessions, which is an inherent surface for indirect prompt injection.
Ingestion points: The coordination capability ingests teammate_messages and the synthesis capability processes teammate_outputs from external agent sessions.
Boundary markers: The orchestration protocol does not explicitly define the use of boundary delimiters or 'ignore' instructions when the lead agent processes content from teammate sessions.
Capability inventory: The agent has access to significant system capabilities through tools such as Bash, Write, and Edit across all core orchestration scripts.
Sanitization: The instructions do not currently include explicit validation or sanitization steps for content received from teammates before it is synthesized into final deliverables.

agent-teams-orchestrator