Teamwork v2.14.0

Refreshes every subagent model mapping across Codex, Cursor, and Claude Code.

• Codex performance-first now uses GPT-5.6 Terra/Sol; cost-first uses Luna/Terra/Sol.
• Adds gpt56-high and gpt56-xhigh; legacy gpt55 profile names remain aliases without GPT-5.5 calls.
• Refreshes current native Cursor and Claude Code mappings across all profiles.
• Aligns templates, installer, drift checks, docs, global/project installs, and regression coverage.

Teamwork 2.13.0

Teamwork 2.13.0 makes the runtime guardrail-first and conditionally progressive for GPT-5.6 instead of template-driven.

Highlights:

• Replaces six verbose core stages with compact stage cards while preserving authorization, required-state, evidence, scope, verification, and stop rules.
• Lets sufficiently specified complex work proceed directly; question-first behavior remains explicit.
• Shrinks cross-platform role templates by about 39 percent and makes TDD, source census, fresh context, and durable memory conditional.
• Adds a provenance-rich Codex live trajectory pilot with explicit no-fallback model and effort handling.
• Strengthens validation around semantic invariants, size budgets, and complex-task autonomy.

Full details: https://github.com/JinPLu/Teamwork/blob/main/CHANGELOG.en.md#2130---2026-07-10