bagbag16

复杂多轮协作守卫。默认轻量使用；当任务跨多轮、易漂移、会积累高影响状态时，展开最小外化与一致性校验，稳住协作同时保持推进。

Formal public entry for Agent Continuity Harness. Use when Codex should keep long-running, drift-prone, cross-window, recovery-sensitive, or stateful collaboration coherent without asking the user to choose internal guard or continuity modes manually. Start in guard-mode by default; enter continuity-mode only when handoff, recovery, formal state, or cross-window continuation is needed.

Evaluate long-term AI/Codex collaboration corpora, chat logs, prompts, revisions, and produced artifacts to build an evidence-grounded cognitive work capability scorecard. Use when the user asks to assess AI collaboration records, abstract a colleague's capability profile from their AI-assisted work, compare collaboration quality across people, or convert long-term chat/productivity traces into neutral dimension scores. Do not use as an IQ test, personality diagnosis, hiring decision automation, or clinical assessment.