The Llm Council

The LLM Council

A multi-model orchestration package that runs adversarial council workflows across OpenAI, Anthropic, Google, Vertex, OpenRouter, and local CLIs.

This is not a Claude-only framework. Claude Code is one supported client and one supported provider path among several.

For tool-driven calls from Codex, Claude Code, Hermes or CI, use the
portable invocation contract.
It covers installed-binary identity, timeouts, process completion, credential
boundaries, complete result files and execution-status handling. The shipped
skill and caller examples require Council 0.8.5 or later for automatic evidence
containment and dependency security floors, in addition to capability-based
Claude admission and the earlier reporting fixes. See the
0.8.8 release notes for the current provider changes
and 0.8.5 release notes for containment details.

Claude Code compatibility is capability-based, not an exact-build allowlist.
Before each request, Council reads the selected binary's version for diagnostics
and checks its help option definitions for the controls the request uses. New
builds can pass this check without a Council release when those definitions
remain available.
Missing controls fail explicitly; Council never drops isolation flags to proceed.
Council captures help through a bounded terminal stream, with a bounded pipe
fallback if a terminal is unavailable. The two permission controls must advertise
the required values; absent choices fail closed. Help output can still be
incomplete in ways this check cannot detect. Help advertises syntax, not
authentication, isolation semantics or generation success. Existing policy
checks and strict terminal-response validation remain.
The Codex provider uses a separately maintained compatibility record
instead of a fixed version…