What problem does this solve?
In auto mode, file writes are applied immediately with no opportunity to review the cumulative changes. Users working on larger refactors want to see a unified diff of all modifications before they are committed to disk.
Proposed solution
Add an agent mode — a high-autonomy mode where lmcode reads, edits, and creates files automatically, then shows a unified diff of all changes for review before finalising.
User flow:
● lmcode (model) [agent] › add docstrings to all functions in src/utils/
⠋ reading src/utils/helpers.py…
⠋ writing src/utils/helpers.py…
⠋ reading src/utils/parsers.py…
⠋ writing src/utils/parsers.py…
─── changes (3 files) ────────────────────
+ src/utils/helpers.py +12 lines
+ src/utils/parsers.py +8 lines
apply changes? [y/n/diff] › diff
[shows unified diff]
apply changes? [y/n] › y
✓ 2 files updated
Key differences from auto mode:
auto applies writes immediately; agent stages them and shows diff firstagent tracks all file writes in a staging area- On completion (or
/done), presents a summary diff and asks to apply
Dependencies:
write_file tool must be implemented (feat/tools-write)- Staging/rollback mechanism for file writes
- Unified diff display (difflib or similar)
Alternatives considered
None considered.
What problem does this solve?
In
automode, file writes are applied immediately with no opportunity to review the cumulative changes. Users working on larger refactors want to see a unified diff of all modifications before they are committed to disk.Proposed solution
Add an
agentmode — a high-autonomy mode where lmcode reads, edits, and creates files automatically, then shows a unified diff of all changes for review before finalising.User flow:
Key differences from
automode:autoapplies writes immediately;agentstages them and shows diff firstagenttracks all file writes in a staging area/done), presents a summary diff and asks to applyDependencies:
write_filetool must be implemented (feat/tools-write)Alternatives considered
None considered.