The Agent Skills Directory

[SAFE]: The skill defines a purely instructional workflow for orchestrating multiple research agents. It does not include any scripts, shell command execution, or network calls to untrusted domains.- [PROMPT_INJECTION]: The skill processes external research data and leaderboard information, creating a surface for indirect prompt injection. Ingestion points: Phase 0 (Research Mission Brief), Phase 3 (Auto-Benchmark), and Phase 4 (Synthesis). Boundary markers: Phase 1.2 explicitly defines thread boundaries and scopes. Capability inventory: Orchestration of Lead Researcher sub-agents and synthesis of research manuscripts. Sanitization: Mitigated by mandatory user-in-the-loop checkpoints before any phase transitions or significant budget reallocations.

principal-scientist