feat(prompt): drop misleading uv pip hint#88
Open
mikasenghaas wants to merge 6 commits into
Open
Conversation
Spell out two things the ipython tool's environment didn't make obvious: - ipython treats `!cmd` per-line; multi-line shell needs `%%bash`. - the rlm tool's uv env is NOT the project venv, so `import <project>` requires sys.path tweak / running under the project interpreter / editable install. Both have cost agents wasted turns on swebench-pro.
Reframe the ipython usage block: - Lead with 'ipython is your scratchpad' (Python state, notes, context). - Make 'cells default to Python' explicit, since the model's prior keeps pushing bare `find ...`/`grep ...` into cells and getting SyntaxError. - List concrete BAD/GOOD antipatterns instead of prose so the rule is easy to pattern-match. - Move the pre-installed package list into this block, drop the 'uv pip install <pkg>' line — that line was misleading on the kernel's uv-tool environment (no .venv -> 'no virtual environment found'). - Drop the 'project venv' commentary; let env-specific append handle it. Also nudge 'breaking down into sub-tasks' in the opening sentence. Validated on rlm-swe SWE-bench Verified n=95 glm-5.1: 64.2 % solve (+6.0 pp vs prior prompt at the same model + dataset).
The conversation log path is internal plumbing — the agent doesn't need to know it, and a system-prompt line per rollout is just per-request token overhead. messages_path is still threaded through the function signature for callers that may use it programmatically.
Drop the bulleted list + BAD/GOOD antipattern table in favor of a single assertive paragraph: frame ipython as the primary working surface / persistent scratchpad, state that cells are Python by default, and fold the shell-escape rule (! / %%bash, bare commands fail) into prose. Keeps the pre-installed-imports note.
…-prompt-notes # Conflicts: # src/rlm/prompt.py
mikasenghaas
changed the title
samsja
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
After merging the IPython control guidance from #90, this PR no longer carries a separate IPython prompt rewrite.
The remaining change removes the misleading system-prompt hint to install packages with
uv pip install <pkg>. That hint points agents toward mutating the IPython kernel environment, which conflicts with the target-runtime guidance now onmain.Verification
uv run ruff check .uv run ruff format --check .uv run pytest tests/test_prompt.pyuv run pytest tests/