[Improvement] Code Editor Tasks

[aaronsmulktis]

Changes

1. add agent Gemma 4 31B
2. add edit_script_add_header_comment task
3. add create_notes_file_in_code_editor task
4. fix sitemap.xml
5. fix register_resolver() error
6. fix editor save method
7. add unit tests

[Copilot]

Pull request overview

Adds first-class “Code Editor” task support by introducing EditFileTask, along with configuration, documentation, and tests to validate reward logic when editing or creating files in the Code Editor app.

Changes:

• Introduces EditFileTask with custom completion logic that inspects the Code Editor file tree directly.
• Adds unit tests and task config entries for editing an existing file and creating a new file.
• Updates docs and a small launcher logging robustness tweak; cleans up a sitemap merge artifact.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/test_codeeditor_tasks.py	Adds unit tests covering EditFileTask reward logic and task instantiation via load_task.
src/open_apps/tasks/tasks.py	Implements EditFileTask and related helpers for matching file content and detecting changes vs initial state.
src/open_apps/tasks/init.py	Re-exports EditFileTask and maps it to the codeeditor app for task filtering.
src/open_apps/launcher.py	Removes local ${now:...} resolver registration and hardens W&B action logging against missing obs/agent_info.
site/sitemap.xml	Resolves merge-conflict artifacts and normalizes sitemap entries.
scripts/conduct_slurm.sh	Updates header comment describing cluster/network assumptions.
docs/tasks.md	Documents EditFileTask usage with YAML examples.
config/tasks/original_tasks.yaml	Adds two Code Editor tasks using EditFileTask.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[marksibrahim]

Thank you for adding the coding editor tasks. These seem reasonable overall. I want to hold off on merging until 1) unit-test pass (currently failing in this branch) 2) I run the task locally on my end. Would you mind looking into 1) for now?

[Copilot]

Pull request overview

Copilot reviewed 16 out of 16 changed files in this pull request and generated no new comments.

Suppressed comments (7)

scripts/conduct_slurm.sh:7

• Spelling/grammar in this comment: “This assume” → “This assumes”, and “Gemme” → “Gemma”.

# target model on :8000), and launches the worker pool pointed at it. This assume
# the eval job is running on the same internal network as Gemme.

config/browsergym_env_args/screenshot.yaml:20

• Typo in comment: “neets” → “needs”.

  base_url: null # this neets to be the localhost url, will be set automatically in launch_experiment.py

config/browsergym_env_args/screenshot.yaml:16

• The inline comment is misleading: headless: True runs without showing a browser window, so it does not “keep the browser open” in a visible sense.

headless: True # keep the browser open

src/open_apps/tasks/tasks.py:732

• Docstring says to provide exactly one of expected_content or required_fragment, but _content_matches() currently allows both (it will silently ignore required_fragment if expected_content is set). This makes misconfigurations easy to miss; consider enforcing mutual exclusivity.

    def _content_matches(self, content: str) -> bool:
        if self.expected_content is not None:
            return self._normalize(content) == self._normalize(self.expected_content)
        if self.required_fragment is not None:
            return self._normalize(self.required_fragment) in self._normalize(content)

src/open_apps/tasks/tasks.py:712

• _normalize() uses .strip(), which removes leading whitespace from the first line and can change the meaning of indentation-sensitive files; the comment above says it only strips trailing whitespace per line. This can make expected_content matches pass/fail incorrectly when the file begins with spaces.

        return "\n".join(line.rstrip() for line in lines).strip()

config/browsergym_env_args/screenshot.yaml:14

• Typo in comment: “envrionments” → “environments”.

This issue also appears in the following locations of the same file:

• line 16
• line 20

task_seed: 42 # not relevant for us we have only deterministic envrionments

docs/tasks.md:100

• The example command references agent=gemma-4, but there is no config/agent/gemma-4.yaml in this repo. Use an existing agent config name so the command is runnable as written.

uv run launch_agent.py agent=gemma-4 task_name=edit_script_add_header_comment

[aaronsmulktis]

Thank you for adding the coding editor tasks. These seem reasonable overall. I want to hold off on merging until 1) unit-test pass (currently failing in this branch) 2) I run the task locally on my end. Would you mind looking into 1) for now?

All failing test are within tests/test_longer_horizon.py which I never touched, after pulling in the latest upstream main I see no changes. Should I debug these longer horizon tests? I will likely require spending some time running those evals and understanding what the tests need to do.