Qin Full Power Mode

Problem

Qin Full Power Mode targets long-running agent tasks where execution can drift: context is collected only once, diagnosis or validation is skipped, risky operations lack confirmation, key decisions get compressed in long conversations, interrupted work is hard to resume, and similar errors repeat. It constrains agent behavior with mandatory phases, status files, and checkpoints.

How It Works

• Activation validation: every activation requires initialization and status setup, so execution does not start from an intermediate state.
• Collect then plan: P0 gathers Git status, history, and workspace structure; P1 selects a strategy by task type, such as debugging, coding, data analysis, or documentation.
• Memory during execution: while P2 performs detailed work, Phase M records important settings, key decisions, user preferences, and project rules in MEMORY.md.
• Risk grading: Phase S treats rm -rf, database deletion, formatting, system-config changes, and service restarts as high-risk operations requiring confirmation or backup.
• Error recovery: Phase E distinguishes fatal, recoverable, and warning-level errors with a fallback path; Phase R creates checkpoints at key steps to support resume.
• Context control: when conversation length or context usage crosses thresholds, it keeps core decisions and important intermediate results while discarding low-value content.

Boundaries

It fits staged engineering tasks that need diagnosis, validation, and recovery, such as troubleshooting, development, ops, data analysis, and research. For one-off simple commands or tasks that need no persistent context, the mandatory workflow adds overhead. It does not replace user confirmation for risky actions, and the source material does not expand every disallowed scenario or decision-tree detail.