Update blueprint to version 3#177
Open
shawn-davis wants to merge 120 commits into
Open
Conversation
* Add first GitHub Actions workflow to test Deploy notebook (#169) * feat: Add first GitHub Actions workflow * Add pull request trigger * fix: update from review commnets --------- Co-authored-by: Eli Fajardo <efajardo@nvidia.com> * Implement email notification for workflow results (#171) Added email notification step to workflow with results. * Add dask config options to config.yml * Concurrency improvements * Update to NAT 1.4 prerelease with async endpoint fix * Update to nvidia-nat 1.4.0 * Update to import to use langchain_classic after NAT upgrade to langchain 1.2.8 * Remove note in README about server log visibility issue for async endpoint since it was fixed in NAT 1.4. * Add explicit dependencies for langchain-classic and filelock * Update the file locking cleanup code to use a non-blocking acquire to avoid the race condition * Guard fetch/checkout when no origin remote exists * Remove commented line from workspace file * Add bokeh dependency for dask dashboard * Add dask dashboard port mapping to docker compose * Add .dockerignore * Mention Dask dashboard in readme --------- Co-authored-by: ryanzhang1230 <ryanzh@nvidia.com>
Made-with: Cursor
Made-with: Cursor
Made-with: Cursor
Made-with: Cursor
**Problem 1: Go binaries not scanned in containers (Dockerfile + package_tools.py)**
`find_go_binaries` relies on `strings` (from binutils) to detect Go binaries
and extract embedded module tables. `binutils` was not installed in the Docker
image, so the tool silently failed — any CVE affecting a Go module compiled
into a container binary (e.g., golang-jwt in etcd) was invisible to the agent.
Changes:
- Add `binutils` to the `apt-get install` block in the Dockerfile so `strings`
and `readelf` are available at runtime.
- Add `_extract_go_modules_via_strings()` fallback in `package_tools.py`: when
`go version -m` is unavailable or times out, parse the embedded Go buildinfo
block directly from the binary using `strings -n 10`. Go statically compiles
a module table with `dep\t<module>\t<version>` lines that are readable as
plain ASCII without the Go toolchain.
- Raise `MAX_FILES` in `find_go_binaries` from 5,000 to 100,000 to handle
large container filesystems (e.g., deepseek-r1 has ~80k files).
**Problem 2: VEX categorizer incorrectly assigning requires_configuration (vex_categorizer.py)**
When the exploitability analyzer returns CONDITIONALLY EXPLOITABLE with a
default-off feature flag as the only barrier, the VEX categorizer was assigning
`requires_configuration` even when: (a) the flag is a documented, user-facing
product feature, and (b) the CVE advisory explicitly marks the installed version
as affected.
Root cause: VEX Step 4 in the decision flowchart directed the agent straight to
`requires_configuration` for any default-off toggle. The override rule
("if the advisory considers the version affected, use `vulnerable`") existed in
the guidance section below but was consistently skipped.
Fix: Surface the override check inline at VEX Step 4 — the exact point where
the agent makes the call — so it cannot be bypassed. No new logic was added;
this duplicates an existing rule at the decision point for visibility.
Validated against CVE-2025-62164 / llm-nim_1.15.2: previously returned
`not_affected / requires_configuration`, now correctly returns
`affected / vulnerable`.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
# Conflicts: # README.md
…FILES_TO_WALK + MAX_GO_BINARIES)
Separate Go binary walk limit from extraction limit (MAX_FILES → MAX_FILES_TO_WALK + MAX_GO_BINARIES) See merge request morpheus/agent-morpheus/vulnerability-analysis!3
Docker Compose and README updates See merge request morpheus/agent-morpheus/vulnerability-analysis!4
Concurrency improvements and remove deprecation warnings See merge request morpheus/agent-morpheus/vulnerability-analysis!6
Update installation instructions in readme See merge request morpheus/agent-morpheus/vulnerability-analysis!8
Update the morpheus labels to match what we think is the golden answer for the agent See merge request morpheus/agent-morpheus/vulnerability-analysis!9
Add RH python eval dataset See merge request morpheus/agent-morpheus/vulnerability-analysis!10
…tries overwrite it if desired. This primarily helps with allowing eval runs to specify the output_dir.
Co-authored-by: Greptile <service_account_group_7407_c52410dfdb7b9ece68d605d180dd8d6a@noreply.gitlab-master.nvidia.com>
…oes-not-work' into 'cve-agent-v3' Made output_dir default come from the workflow while letting input entries... Closes #6 See merge request morpheus/agent-morpheus/vulnerability-analysis!11
Update shipped NVIDIA endpoint defaults from Inference Hub to Integrate, replace GPT 5.6 defaults with Nemotron 3 Super, and move the subagent LLM key/default to Nemotron 3 Nano. Refresh README, env example, deploy compose defaults, and the CVE customization notebook to match the new hosted model names.
Classify model failures before retrying so context-window, auth, missing-model, and bad-request errors fail fast while transient rate-limit, timeout, and server failures remain retryable. Add plugin tests for the retry classifier and middleware wiring.
Decouple successful stage reuse from output_dir so equivalent requests share work while materializing reports into their requested output group. Canonicalize presentation paths, expose hit/miss diagnostics, and isolate analysis policies with explicit fail-closed cache namespaces. Hash every semantic container input, including image-config content, and preserve cached phase metadata instead of inferring it on a hit. Existing output reports are intentionally not imported, so the first run after upgrade is cold. No prompts, analysis policy, or dependencies change.
…ss-process file-slot lock
…extraction concurrency at startup
…ss locking holds across Dask workers
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
5 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
This PR brings the
aiq_integrationbranch intoupstream/mainand updates the project around the current NAT-based vulnerability-analysis workflow.Key changes include:
cve_pipelineNAT workflow with structuredPipelineInput/PipelineResultmodels.crane, including caching and a standalone extraction helper.deploy/, refreshes deployment notebooks, and adds containerized quick-start guidance..env.example, documents submodule initialization,cranesetup, API key checks, and common troubleshooting.Validation
Performed during this branch cleanup:
git diff --checkpython3 -m py_compile src/vuln_analysis/eval/sample_pb_eval.py/raid/...or/cyberpub/sdavis/vulnerability-analysis/...paths remain.Notes
This is a large integration PR. Review is probably easiest by area: pipeline/data models, agent implementations, container extraction, deployment/docs, and eval tooling.
After checkout, run:
Local automatic image extraction also requires
craneonPATH.