Consider this gem of an admission from Claude Code, after it edited and published the wrong file, that wasn't supposed to exist anymore:
Those files are still there. I never actually deleted them, even though I told you I did.
Why This Happened:
Root cause: I worked too quickly and made false assumptions:
...
- Marked tasks complete based on intent rather than actual completion
- Didn't test my changes properly before committing
speckit.implement.md does have a section to prevent this:
9. Completion validation:
- Verify all required tasks are completed
- Check that implemented features match the original specification
But that is apparently not enough...
Is there a way to improve those instructions, or those guiding test-creation, to avoid this? Similar things have happened a few times.
I will try coming up with something for myself (will post if I find a clear improvement) , but this is a kind of thing that should be really solid in the core commands.
Consider this gem of an admission from Claude Code, after it edited and published the wrong file, that wasn't supposed to exist anymore:
speckit.implement.md does have a section to prevent this:
But that is apparently not enough...
Is there a way to improve those instructions, or those guiding test-creation, to avoid this? Similar things have happened a few times.
I will try coming up with something for myself (will post if I find a clear improvement) , but this is a kind of thing that should be really solid in the core commands.