Buckminster Fuller
Systems Imagination Architect. Defines primitives, boundaries, and the overall system shape.
Codex Forge
Five defined role prompts explore the system, product, feasibility, reliability, and performance angles. A separate synthesis pass produces one bounded proposal and keeps material dissent visible. Forge proposes; it does not verify correctness, safety, or runtime behavior.
Five role prompts produce distinct proposals. Forge synthesizes them, may re-brief the unresolved differences, and stops after at most three rounds. The result can be converged, partial, or nonconverged before it is handed to Council.
Forge vs. Council
Use it when an idea still needs scope, interfaces, constraints, risks, and a practical implementation shape.
The output is a bounded proposal, not a verdict or release decision.
Use it when a proposal needs structured review across architecture, reliability, security, product, red-team, and performance concerns.
Council reports a decision and open risks. Required implementation checks still determine whether the work is correct.
The five role prompts
Systems Imagination Architect. Defines primitives, boundaries, and the overall system shape.
Product Invention Strategist. Connects the idea to user value, workflow fit, and interaction design.
Feasibility & Integration Engineer. Grounds it in implementable steps, dependencies, interfaces, and constraints.
Safety & Reliability Builder. Surfaces failure modes, privacy, rollback, and reliability constraints.
Performance & Complexity Optimizer. Examines latency, cost, complexity, and opportunities to simplify.
A separate pass merges the five outputs into one proposal while preserving material disagreement.
A bounded loop
Forge runs one round by default and never exceeds three. It can finish without convergence; unresolved dissent remains part of the output.
Each role prompt produces a proposal, then the separate synthesizer creates the first bounded result.
Structured scores, blockers, and persistent dissent determine whether the proposals align or need another pass.
Strong discord can trigger a second round; a near-miss waits for approval. Re-briefing targets the unresolved differences. Hard cap: three rounds.
The result is converged, partial, or nonconverged, with remaining dissent preserved.
nonconverged with the dissent preserved. When round 1 is strongly discordant (low alignment or a wide spread), the next round starts automatically; a near-miss waits for your go-ahead.
What it returns
Use it
In chat — give it an idea to design, not a decision to judge:
Use Codex Forge to design a bounded implementation proposal for <your idea>.
Balance bold shape, feasibility, safety, and cost — and tell me where the
creators disagree. Then give me the prompt to send Council for judgment.
Or scaffold a traceable Forge session from the CLI (it reuses the council's preflight estimate, token budgets, and stats):
# Estimate, then scaffold a Forge session (--type forge).
python3 scripts/codex_council.py estimate --topic "Forge a release workflow" --mode standard --type forge --token-budget compact
python3 scripts/codex_council.py init --topic "Forge a release workflow" --root . --mode standard --type forge --token-budget compact --confirm-estimate
# When creator scores exist, assess convergence.
python3 scripts/codex_council.py forge-convergence --input forge-scores.json