full-mode-orchestrator

name	full-mode-orchestrator
description	Orchestrate complex coding or analysis work through the workspace Full Mode framework (role cards, gates, presets, subagent plans, and release judgment). Use when a task needs multi-role execution, objective PASS/FAIL gating, specialist preset selection, role coverage validation, or patch-loop iteration before release.

Full Mode Orchestrator

Run work through framework/full-mode with explicit roles, gates, and verifiable outputs.

Quick Start

1. Validate role library and preset integrity:
   • python3 framework/full-mode/scripts/validate-role-coverage.py
2. Pick preset (software-low, ml-prod, research-deep, debug-root-cause, architecture-split).
3. Run task:
   • python3 framework/full-mode/runtime/run_task.py --preset <preset> --task-file <TASK.md> --emit-subagent-plan
4. Review report + blockers in framework/full-mode/runtime/runs/<run-id>/report.json.
5. Run patch loop until final decision is PASS.

Workflow Rules

• Never skip Intake, Verification, or Release gates.
• Treat Release Judge output as final gate decision.
• If architecture or scope drifts during implementation, restart from Architecture stage.
• For DS/ML work, enforce data/eval/perf/security checks before release.

Preset Selection

• software-low: normal software feature with core checks.
• ml-prod: production ML/AI with full data+eval+ops rigor.
• research-deep: structured research and synthesis workflow.
• debug-root-cause: incident/regression diagnosis then patch verification.
• architecture-split: parallel backend/frontend architecture alignment.

If no preset fits, clone nearest preset and keep mandatory gates.

Role Policy

• Use canonical roles from roles/ROLE-MATRIX.md.
• Alias roles are compatibility-only; prefer canonical roles in new presets.
• Require role outputs to include blockers and clear ownership.

References

• Role index: references/role-library-map.md
• Runtime execution cheatsheet: references/runtime-ops.md