Skip to content

fix(agent): say what was done when the step limit is hit - #82

Merged
Rl0007 merged 1 commit into
mainfrom
fix/agent-round-limit-summary
Oct 6, 2026
Merged

Rl0007 merged 1 commit into
mainfrom
fix/agent-round-limit-summary

Conversation

@Rl0007

@Rl0007 Rl0007 commented Oct 5, 2026

Copy link
Copy Markdown
Contributor
  1. When the assistant ran out of steps in the middle of a big request (for example, moving many sections), it only said "took too many steps". Some sections had moved and some had not, and it did not say which.
  2. Now it says that it stopped early, lists each change it made in this turn, and asks you to tell it to continue for the rest.

You can see it in the assistant panel on any import.

Fixes #28

After

Verification
  • New test_round_limit_reports_what_was_done in test_agent_write.py: it fails on main (the message was the generic "took too many steps" text) and passes with this fix. The LLM is stubbed.
  • Agent test modules all pass on the devbox. The full suite gives 486 tests: 475 pass, 11 skipped, 0 failures (baseline 485 plus the new test).
  • Devbox e2e on a [test] import with a stubbed LLM (no model on the box): 3 real moves, 1 lookup, 1 failed move, then the step limit. The final message lists all four write results. The moves were undone afterwards.
  • The red line above the input now shows only the first line, so the list does not appear twice.
  • Reviewer findings were applied: clearer wording, a stronger test and a shorter error line.

@greptile-apps greptile-apps Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Your trial has ended. Reactivate Greptile to resume code reviews.

@Rl0007
Rl0007 merged commit 07bf936 into main Oct 6, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Moving many sections with the assistant stops half done without saying what moved

1 participant