The Agent Skills Directory

Agent Orchestration and Model Pinning: The skill uses a 'council' architecture where it spawns multiple subagents, each pinned to a specific model (such as GPT-5.5 or Claude 4.6). This is a standard method for gathering diverse perspectives and reducing single-model bias during complex planning tasks.
Controlled Research Environment: The instructions explicitly constrain subagents to a read-only research mode, directing them to stay within scope and avoid suggesting patches or speculating. This minimizes the risk of unintended code modifications during the planning phase.
Structured Data Processing: The workflow includes a 'Phase 2' where a planning brief is constructed from gathered context. This brief acts as a structured input for the subagents, ensuring they have consistent baseline information while being encouraged to verify details independently using standard codebase tools.
Consensus and Synthesis Logic: The skill provides detailed logic for synthesizing independent proposals into consensus approaches, alternatives, and risks. This process includes safeguards to drop over-engineered or speculative proposals, focusing the final output on high-confidence insights.

council-plan