1. Why Do You NEED This Feature?
Allowing you to create hooks towards certain goals will allow more elaborate harness engineering that would help the model execute their task better. Just as example of what an agent did and how it responded why it didn't do better.
The tool requires exact matching of old_string to existing file content. That works fine for small surgical edits. The problem arose because I was trying to replace a very large block of text (a whole test method that was 30 lines). I did not want to type the entire replacement at once because it was long and I worried about getting whitespace wrong, so I tried to "build up" the new string incrementally -- replacing a few characters at a time, growing the replacement outward.
That was the mistake. The tool itself is fine. What I should have done instead was:
• Read the file
• Draft the complete old block and new block in my head
• Submit one single old_string / new_string pair with the full method replaced
The incremental approach was like trying to paint a house one brush-hair at a time instead of just rolling the whole wall. The tool wanted me to give it the full canvas, and I kept insisting on atomizing it.
That last comment was particularly funny, but I digress. Giving the agent more signals that it could take a step back and consider the larger situation would yield better results. Forgecode currently lacks this, and it be pretty great if there were ways to inject those signals.
2. What Is NOT Possible Right Now?
See above.
3. What WILL Be Possible With This Feature?
See above.
Proposed Solution (User Experience)
It's in there.
Alternatives Considered
No response
Feature Category
Performance
Priority/Impact
High - Would significantly improve my workflow
Examples from Other Tools
No response
Additional Context
No response
Pre-submission Checklist
1. Why Do You NEED This Feature?
Allowing you to create hooks towards certain goals will allow more elaborate harness engineering that would help the model execute their task better. Just as example of what an agent did and how it responded why it didn't do better.
That last comment was particularly funny, but I digress. Giving the agent more signals that it could take a step back and consider the larger situation would yield better results. Forgecode currently lacks this, and it be pretty great if there were ways to inject those signals.
2. What Is NOT Possible Right Now?
See above.
3. What WILL Be Possible With This Feature?
See above.
Proposed Solution (User Experience)
It's in there.
Alternatives Considered
No response
Feature Category
Performance
Priority/Impact
High - Would significantly improve my workflow
Examples from Other Tools
No response
Additional Context
No response
Pre-submission Checklist