security: bump nltk >=3.10.0 and fix inisec import blocks - #2290
Merged
Conversation
Bump nltk>=3.10.0 in iheval and rolemrc (CVE fix), and pre-import regex and defusedxml.ElementTree in all servers whose import chains reach nltk. nltk>=3.9 added NLTKSafeImportFinder (inisec.py) which blocks imports triggered by nltk if the module path falls inside the process CWD. Server venvs live inside the repo root in CI, so legitimate site-packages get blocked. Pre-importing before the nltk chain fires puts packages in sys.modules and bypasses the check. ifbench also needs the pre-imports in conftest.py so _ensure_nltk_data() can download punkt without being silently blocked during pytest_configure. Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Kajal Jain <kajalj@nvidia.com>
Signed-off-by: Kajal Jain <kajalj@nvidia.com>
Signed-off-by: Kajal Jain <kajalj@nvidia.com>
thomasdhc
approved these changes
Aug 3, 2026
kajalj22
added
the
r0.5.0
bxyu-nvidia
approved these changes
Aug 3, 2026
hemildesai
added a commit
that referenced
this pull request
Aug 3, 2026
Picks up the nltk bump that unblocks the ifbench/iheval/rolemrc/toolsandbox and instruction_following server suites (#2290). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Hemil Desai <hemild@nvidia.com>
cmunley1
added a commit
that referenced
this pull request
Aug 3, 2026
## Summary
\`test_sanity_query_input\` was failing in CI because \`review_model\`
is present in \`test_data.json\` (the snapshot) but absent from the
actual server response in some tau2-bench versions.
\`review_model\` was added to \`TextRunConfig\` in tau2-bench v1.0.1
with a default of \`'claude-opus-4-5'\`. Its presence in the serialized
config depends on which tau2-bench commit is installed. The test's
\`_clean()\` function already strips other volatile config fields
(\`max_agent_steps\`, \`turns_remaining_interval\`) — adding
\`review_model\` follows the same pattern and makes the comparison
stable across tau2-bench versions.
## Change
One line added to \`_clean()\` in
\`responses_api_agents/tau2/tests/test_app.py\`:
\`\`\`python
d["config"].pop("review_model", None)
\`\`\`
## Verified
6/6 tau2 tests pass locally.
Failure first surfaced in the full suite run off #2290:
https://github.com/NVIDIA-NeMo/Gym/actions/runs/30840804435
Signed-off-by: Kajal Jain <kajalj@nvidia.com>
Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Christian Munley <cmunley@nvidia.com>
kajalj22
added a commit
that referenced
this pull request
Aug 4, 2026
…` into `r0.5.0` (#2291) beep boop [🤖]: Hi @kajalj22 👋, we've cherry picked #2290 into for you! 🚀 Please review and approve this cherry pick by your convenience! Signed-off-by: Kajal Jain <kajalj@nvidia.com> Signed-off-by: NeMo Bot <nemo-bot@nvidia.com> Co-authored-by: kajalj22 <kajalj@nvidia.com> Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
OlegSudakov
pushed a commit
to OlegSudakov/Gym
that referenced
this pull request
Aug 7, 2026
…o#2290) ## Summary Bump \`nltk>=3.10.0\` in \`iheval\` and \`rolemrc\` (CVE fix), and pre-import \`regex\` and \`defusedxml.ElementTree\` in all servers whose import chains reach nltk. ## Root cause \`nltk>=3.9\` added \`NLTKSafeImportFinder\` (\`nltk/inisec.py\`) which blocks any import triggered by nltk if the module's resolved path is inside the process CWD. In CI, server venvs live at \`resources_servers/<name>/.venv/\` — inside the repo root — so legitimate site-packages get blocked. Pre-importing before the nltk chain fires puts packages in \`sys.modules\`, bypassing the check. \`ifbench\` also needs the pre-imports in \`conftest.py\` so \`_ensure_nltk_data()\` can download punkt without being silently blocked during \`pytest_configure\`. ## Changes | File | Change | |---|---| | \`resources_servers/{iheval,rolemrc}/requirements.txt\` | bump \`nltk>=3.10.0\` | | \`resources_servers/{iheval,rolemrc,ifbench,instruction_following,toolsandbox}/app.py\` | pre-import \`regex\`, \`defusedxml.ElementTree\` | | \`resources_servers/ifbench/tests/conftest.py\` | same pre-imports before \`ensure_ifbench()\` | ## Local test results | Server | Result | |---|---| | iheval | 88/88 ✓ | | rolemrc | 55/55 ✓ | | ifbench | 15/15 ✓ | | instruction_following | 15/15 ✓ | | toolsandbox | 19/19 ✓ | ## Full test suite run https://github.com/NVIDIA-NeMo/Gym/actions/runs/30840804435 --------- Signed-off-by: Kajal Jain <kajalj@nvidia.com> Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
OlegSudakov
pushed a commit
to OlegSudakov/Gym
that referenced
this pull request
Aug 7, 2026
…#2292) ## Summary \`test_sanity_query_input\` was failing in CI because \`review_model\` is present in \`test_data.json\` (the snapshot) but absent from the actual server response in some tau2-bench versions. \`review_model\` was added to \`TextRunConfig\` in tau2-bench v1.0.1 with a default of \`'claude-opus-4-5'\`. Its presence in the serialized config depends on which tau2-bench commit is installed. The test's \`_clean()\` function already strips other volatile config fields (\`max_agent_steps\`, \`turns_remaining_interval\`) — adding \`review_model\` follows the same pattern and makes the comparison stable across tau2-bench versions. ## Change One line added to \`_clean()\` in \`responses_api_agents/tau2/tests/test_app.py\`: \`\`\`python d["config"].pop("review_model", None) \`\`\` ## Verified 6/6 tau2 tests pass locally. Failure first surfaced in the full suite run off NVIDIA-NeMo#2290: https://github.com/NVIDIA-NeMo/Gym/actions/runs/30840804435 Signed-off-by: Kajal Jain <kajalj@nvidia.com> Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Christian Munley <cmunley@nvidia.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
3 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
Bump `nltk>=3.10.0` in `iheval` and `rolemrc` (CVE fix), and pre-import `regex` and `defusedxml.ElementTree` in all servers whose import chains reach nltk.
Root cause
`nltk>=3.9` added `NLTKSafeImportFinder` (`nltk/inisec.py`) which blocks any import triggered by nltk if the module's resolved path is inside the process CWD. In CI, server venvs live at `resources_servers//.venv/` — inside the repo root — so legitimate site-packages get blocked. Pre-importing before the nltk chain fires puts packages in `sys.modules`, bypassing the check.
`ifbench` also needs the pre-imports in `conftest.py` so `_ensure_nltk_data()` can download punkt without being silently blocked during `pytest_configure`.
Changes
Local test results
Full test suite run
https://github.com/NVIDIA-NeMo/Gym/actions/runs/30840804435