deliberate

Deliberate

Contract

• Use a Coordinator, a Synthesizer, and an independent Challenger. The Coordinator may synthesize, never challenge.
• Route exchanges through the Coordinator; reuse both roles and send complete state.
• Prefer real isolated subagents for the Synthesizer and Challenger when the host can dispatch them; fall back to inline role-play with ISOLATION: simulated disclosed. Follow Subagent Dispatch.
• Discuss internally first. Ask only about material user-owned choices that research, reasoning, or a reversible default cannot settle.
• Process every related topic and visible issue per pass; never drip-feed known issues.
• Set no round limit. Use CONTINUE only for a material open issue with a concrete next-pass change; otherwise stop or pause.
• Without any Challenger role, disclose that the discussion cannot run. In INLINE-FALLBACK, a simulated Challenger may return SUFFICIENT only with ISOLATION: simulated disclosed; never present simulated convergence as independent sufficiency.
• Persist every delivered result under the owning task's tasks/artifacts/<task-id>/discussions/; create other resources only when authorized. If the owning canonical task ID is unavailable, return to task activation before writing.

State

Have the Coordinator maintain and transmit:

QUESTION AND GOAL:
CONSTRAINTS / NON-GOALS:
FACTS / SOURCES:
TOPIC BUNDLE:
ASSUMPTIONS / DEFAULTS:
CURRENT ANSWER:
ISSUES / RESOURCES:
LIMITATIONS / USER DECISIONS:

Give each topic a fixed T-ID. Allocate D-IDs monotonically; never reuse or renumber them, and reopen an issue under its original ID. Send the complete answer, not a diff.

Follow Resource Handoff. Verify access; changed resources invalidate dependent conclusions.

Workflow

0. Resolve dispatch mode

Detect the host's dispatch capability, verify the dispatch path is actually ready (on Pi: the pi-agent carrier is registered and spawnable; on Codex: the host can spawn an agent process), record the mode and per-role readiness once per run, and print the dispatch metadata table to the user as the first output of the run — before any role pass — as specified in Subagent Dispatch. If all roles are ready, enter SUBAGENT mode without asking; if any role is missing, present the table and ask the user whether to proceed in INLINE-FALLBACK, and enter it only on explicit confirmation. Never enter the loop on an unverified dispatch path. In SUBAGENT mode, run the Synthesizer and Challenger as isolated subagents for every pass; in INLINE-FALLBACK mode, run them inline with ISOLATION: simulated. The state packet, D-ID/T-ID ledger, and CONTINUE/SUFFICIENT rules are identical in both modes.