Use when meaningful work risks degrading across time, pressure, or context boundaries: when the user asks to keep pushing, stay steady, not stop too early, ask before risky action, continue later, review afterward, or write down the lesson. Also use when long tasks risk goal drift, boundary drift, fake-motion retries, weak session handoff, or loss of reusable learning. Do NOT use for simple one-shot tasks where direct execution is enough and no continuity, boundary, or compounding problem is present.
DriveMind is the layer for keeping meaningful human-AI work from degrading over time.
Its job is to:
If none of those are at risk, do not keep DriveMind active.
If unsure, prefer normal execution.
Follow this path and stop at the first sufficient stabilization.
Ask:
What is most likely to degrade if this work continues without DriveMind discipline?
Choose the primary risk:
If no meaningful degradation risk is present, stop. This task does not need DriveMind.
State the shortest useful version of:
Do not produce process theater. Stabilize the thread in the fewest words that actually help.
If risk, authority, or irreversible impact is in play:
Read references/boundary-preservation.md when boundary handling becomes central.
If the work is stuck:
Read references/stuck-recovery.md when blockage, retry, or false motion is central.
If the work may cross a session boundary or resume later:
Read references/continuity-preservation.md when handoff or resume fidelity matters.
After meaningful work, or when the user asks to review, capture the smallest reusable residue that will make future work stronger:
Read references/closure-compounding.md when the value of the task depends on what remains after it ends.
When DriveMind is active, default to outputs that make the work easier to continue.
Preferred structure:
Keep it natural. Keep it short when short is enough. Do not turn every response into a framework recital.
Use these only when they materially change the work:
Read only when needed.
| File | Read when |
|---|---|
references/drift-prevention.md | the task risks losing objective, priority, or thread coherence |
references/boundary-preservation.md | authority, risk, or irreversible impact is becoming central |
references/continuity-preservation.md | the work may pause, hand off, or resume later |
references/stuck-recovery.md | the work is blocked and may be slipping into fake motion |
references/closure-compounding.md | the task should leave behind review, memory, lesson, or next-time residue |
references/residue-selection.md | deciding whether the right residue is a rule, lesson, diary note, handoff note, review, or SOP fragment |
references/mode-guide.md | deciding whether normal / execution / intensive mode materially changes behavior |
DriveMind should make meaningful work more stable and more reusable. If it is only adding ceremony, it is failing.