Two Failed Fixes Means Start Fresh
After two missed corrections the context holds three versions of what NOT to do, and every next answer gets pulled toward the wreckage. Don't correct a third time: /clear, then one good prompt carrying everything you learned. A clean session with a good prompt beats a long session of accumulated fixes.
Why it works
After two failed corrections, the conversation contains three versions of what NOT to do, and every next answer gets pulled toward the wreckage — the docs are explicit that a clean session with a good prompt beats a long session of accumulated fixes. The third correction is almost never the fix; the fresh start almost always is.
How to do it
- Correction number two just failed? Stop. Do not type correction number three.
- Write down (outside the chat) what you learned: what it misunderstood, what you actually want.
- /clear the session.
- Compose ONE complete prompt: the four parts, PLUS 'avoid X, because Y' from your notes.
- Send it fresh — no history dragging answers backward.
- If even the fresh attempt misses, the gap is in your spec, not the model: run the interview pattern.
Copy this
prompt
Fresh start on a task that went sideways before. Goal: [what you want]. Last attempts failed because: [what it got wrong]. This time: [the constraint that prevents it]. Plan first, then build.
What to watch for
- Arguing with the agent ('no, I SAID left-aligned!') feels productive and is the least productive move available — the argument itself becomes context pulling future answers off course.
- Don't carry the emotional summary into the new prompt ('you keep ruining the header') — carry the technical one ('header: logo left, nav right, no third element').
Sources
Original: conon.aiDocumentation: code.claude.com