Remove RLM_DEFAULT_TOOL_NAMES, accept rlm_tools#1223
Merged
Conversation
Lets callers declare which builtin RLM tools are active so
ToolMonitorRubric tracks exactly that set. Unset keeps today's default
(ipython + summarize). Callers are expected to pass the same list via
ComposableEnv(environment_vars={"RLM_TOOLS": ...}) so the RLM sandbox
advertises the same tools; this pairing is done at the research-env
level.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lets a harness own sandbox env vars that are logically paired with other harness config, so they can't silently desync. Use case: rlm_harness(rlm_tools=[...]) now sets both Harness.tool_names (for ToolMonitorRubric) and Harness.environment_vars["RLM_TOOLS"] (for the sandbox) from the same single kwarg. Research envs stop maintaining a parallel "RLM_TOOLS": ",".join(...) entry in their own environment_vars dicts. Merge order in ComposableEnv.build_env_vars (first→last, later wins): caller-supplied environment_vars → harness.environment_vars → taskset.get_env_vars() → AGENT_WORKDIR. Harness wins over caller by design so a stray caller override can't break a harness-level pairing. Harness keys collide-check against PROTECTED_ENV_VARS + AGENT_WORKDIR, same as taskset. Also adds docstring to rlm_harness making the fan-out explicit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit f14048b. Configure here.
The Harness field enumeration in docs/environments.md had drifted from the dataclass. Adds the 5 missing fields in declaration order: sandbox_spec, skills_path, get_upload_dirs (pre-existing omissions), tool_names (pre-existing), and environment_vars (this PR). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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
1 participant
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.
Description
The
rlm_harnesstakesenvironment_varsand theComposableEnvmerges those with its own. That way, we have a unified way to drive both sandbox behavior and verifiers behavior from the harness, while still allowing for environment vars that are given toComposableEnv.rlm_harnessnow takes theRLM_TOOLSenvironment vars.Type of Change
Testing
uv run pytestlocally.Checklist
Additional Notes
Note
Medium Risk
Changes how sandbox environment variables are merged for
ComposableEnvand introduces harness-owned env vars, which could alter runtime behavior for existing harness/tasksets if keys collide or ordering assumptions existed.Overview
ComposableEnvnow merges harness-provided sandbox environment variables (Harness.environment_vars) into the sandbox env, with validation preventing overrides of protected keys.rlm_harnessis updated to accept anrlm_toolslist and uses it as the single source of truth for both tool monitoring (tool_names) and the sandbox’sRLM_TOOLSenv var, removing the prior fixed default constant. Documentation is updated to reflect the expandedHarnessconfig surface.Reviewed by Cursor Bugbot for commit dcb4aa1. Bugbot is set up for automated code reviews on this repo. Configure here.