Core idea
The default council works immediately after the skill is installed. It spawns multiple role-based sub-agents, but all of them run on the same underlying Claude model. This mode provides multi-agent deliberation without requiring an external multi-model service or a separate API key.
How it works
The user invokes the LLM Council skill and supplies the decision question. The default workflow then creates five sub-agents. Each agent has its own prompt, and the agents participate in the stochastic debate and review process.
Although the roles are separate, every sub-agent in this mode uses the same model. In one run, that model is an Opus model. The user can open the individual agent entries to view their prompts and outputs as the run progresses.
The default mode therefore provides:
- Multiple agents.
- Different role prompts.
- Independent agent outputs.
- Review and final synthesis.
- One shared underlying model.
Why it matters
This mode is operationally simple. Once the skill is installed, it works without connecting another service. It still improves on a direct answer by forcing the shared model to approach the decision through several configured roles.
The operating recommendation is that this mode is good enough most of the time. It provides a practical baseline for users who want a council process without the added configuration and usage charges of the cross-vendor workflow.
Practical application
Invoke /LLM Council and follow it with a complete strategic brief. For example, provide:
- The two business paths being considered.
- The revenue target.
- The current monthly recurring revenue of the existing offer.
- What additional work is required to scale that offer.
- The nature of the alternative service business.
- A direct request for the council to recommend which path should receive more focus.
After launching the run, inspect the agents as they appear. The presence of several named agents and separate prompts confirms that the council is operating rather than returning an ordinary single response.
Trade-offs and limitations
All agents inherit the same underlying model's tendencies. Prompt diversity can produce different perspectives, but it does not eliminate vulnerabilities associated with using only one model family.
The default mode can also reach a materially different conclusion from the cross-vendor mode. In the business example, the single-model council favored scaling the existing membership, while retaining high-ticket work as a secondary cash bridge and lead source. That recommendation should be understood as the output of one council configuration, not as an architecture-independent result.
Key takeaway
Use the default council as the low-friction operating mode: several prompted agents, one underlying model, no additional multi-model service, and enough deliberation for most uses described in this setup.