The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection as its core function involves ingesting and synthesizing data from untrusted external sources like web searches, GitHub repositories, and academic papers.
Ingestion points: Untrusted data enters the agent context through research subagents assigned to external search and exploration tasks (documented in references/anti-slop-research.md).
Boundary markers: The skill employs structured 'Prompt Packets' to strictly delimit subagent objectives and a 'Synthesis Gate' to filter and validate findings before they are integrated into the lead agent's context (references/prompt-packets.md, references/synthesis-gate.md).
Capability inventory: The skill has the capability to spawn subagents and integrate their results, which may include filesystem modifications if authorized by the host harness (references/harness-adapters.md).
Sanitization: It incorporates 'Anti-Slop Research' protocols designed to identify and discard adversarial, low-quality, or AI-generated content, while mandating that every integrated claim is supported by direct evidence or artifacts (references/anti-slop-research.md).
[EXTERNAL_DOWNLOADS]: The skill references technical guidelines and documentation from established technology organizations.
Evidence: The skill includes references to official documentation and best practices hosted by OpenAI and Anthropic for model-routing and subagent configuration (references/source-notes.md).

orchestrate-subagents