The Agent Skills Directory

[SAFE]: The skill implements a secure-by-design deliberation protocol that emphasizes role isolation and context management. In Phase 2, agents are instructed to provide contributions without seeing others' outputs, which prevents consensus-based prompt injection or 'groupthink' bias. The use of a two-layer memory system (raw transcripts vs. approved decisions) ensures that only founder-validated data is used for future context, preventing hallucinated or unauthorized data from influencing the agent's long-term behavior.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it processes user-provided topics and may perform external research. This is managed by the protocol's structure which mandates that all synthesized decisions undergo a manual 'Human in the Loop' review before being recorded to the permanent decision memory.\n
Ingestion points: User-defined strategic topics via the /cs:board command and external research data acquired during the C-suite contribution phase (SKILL.md).\n
Boundary markers: The protocol uses distinct phases, role-specific contribution blocks, and markdown templates to separate different streams of information.\n
Capability inventory: The skill is capable of writing logs and decision files to a local memory/ directory (SKILL.md).\n
Sanitization: While explicit sanitization of research data is not detailed, the requirement for human approval acts as a primary control against malicious inputs affecting the system's output.

board-meeting