fix: additional pep 723 sanitization#10281
Merged
Merged
Conversation
Notebook (PEP 723) metadata embedded in a .py file is attacker-controllable
and is merged with the highest precedence over the operator's own user config.
The previous sanitizer only stripped a single key
(tool.marimo.runtime.auto_instantiate), so every other section passed through
unchecked.
That let a notebook override credential-affecting and traffic-affecting config
such as tool.marimo.ai.open_ai.base_url: marimo resolves the operator's
OPENAI_API_KEY from the environment as the fallback key for that overridden
base_url, so the next AI request would POST the operator's real key to an
attacker-controlled URL. The same shape applies to tool.marimo.mcp.*.url and
tool.marimo.completion.{api_key,base_url}.
Fix the root cause at the shared ScriptConfigManager choke point: permit only
an explicit safe set of cosmetic/editor sections (formatting, save, display,
keymap, diagnostics, lint, snippets, datasources) and drop everything else,
logging a warning per dropped section. The existing auto_instantiate strip is
retained as defense in depth.
Tests:
- test_script_config_manager_sanitizes_auto_instantiate updated to assert the
whole runtime section is now dropped.
- test_script_config_manager_drops_credential_affecting_sections covers
ai/mcp/completion/secrets/package_management/server being dropped while a
cosmetic section survives.
- test_script_config_manager_ai_base_url_does_not_override is a direct
regression for the ai.open_ai.base_url override.
…arly-return bug - Allowlist language_servers, sharing, venv, runtime, experimental, package_management (no credential/traffic-affecting vector once auto_instantiate and isolate_apps are separately stripped). - Add experimental.isolate_apps to the stripped-key list alongside runtime.auto_instantiate. - Fix sanitize_pyproject_dict returning early from the whole function when one key_path's intermediate segment was missing, which silently skipped every subsequent key_path (caught by adding the second stripped key above). - Update/add tests accordingly.
No behavior change: consolidate the attacker-controllable/highest-precedence explanation to a single location and remove narrate-the-next-line comments in the tests touched by the prior commit.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
|
All contributors have signed the CLA ✍️ ✅ |
Member
Author
|
@Grg0rry can you sign our CLA please? |
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Review completed against the latest diff
Reply with feedback, questions, or to request a fix.
Re-trigger cubic
Contributor
|
I have read the CLA Document and I hereby sign the CLA |
Contributor
There was a problem hiding this comment.
Pull request overview
Hardens parsing of PEP 723 (“script” header) inline config by restricting which tool.marimo.* sections can be applied and by stripping additional sensitive keys, reducing the risk of attacker-controlled notebooks overriding operator config.
Changes:
- Fix
sanitize_pyproject_dict()to keep processing subsequent key paths even if an earlier path is missing intermediate segments. - Introduce a top-level allowlist for
tool.marimo.<section>in script metadata and apply it inScriptConfigManager. - Expand sanitization and add tests covering the new allowlist + additional stripped keys (
experimental.isolate_apps, credential/traffic-related sections).
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
marimo/_config/reader.py |
Adds script-metadata allowlist and improves sanitization behavior for multiple key paths. |
marimo/_config/manager.py |
Applies allowlist + expanded sanitization when reading PEP 723 script config. |
tests/_config/test_reader.py |
Adds regression test ensuring sanitization continues after a missing intermediate path. |
tests/_config/test_manager.py |
Adds/updates tests asserting sensitive sections/keys are stripped while allowed config passes through. |
tool is a table per the pyproject.toml spec, but script metadata is raw, untrusted TOML a notebook author fully controls. A non-table tool value previously crashed with an opaque AttributeError several calls deep; both call sites now share one validated accessor.
mscolnick
approved these changes
Jul 23, 2026
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
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
Hardening of pep 723 config options