Public scan — anyone with this URL can view this analysis. Sign up to track your own repos privately, run scheduled re-scans, and get AI fix prompts via your dashboard.
109 of your 217 findings came from Repobility's proprietary detections. ✓ Repobility tags below mark them.

Scan timing: clone 2.89s · analysis 9.79s · 15.3 MB · GitHub API rate-limit (preflight)

ai-dynamo/nixl

https://github.com/ai-dynamo/nixl · scanned 2026-06-05 20:03 UTC (4 days, 14 hours ago) · 10 languages

284 raw signals (200 security + 84 graph) 45th percentile · Cpp · medium (20-100K LoC) System graph score 84 (lower by 19)

UNIFIED Repobility · multi-layer engine · AI coders

Complete repo analysis

Last scanned 4 days, 14 hours ago · v2 · 97 actionable findings from 2 signal sources. 132 repeated signals grouped for readability. Security checks, system graph analysis, and verified AI-agent feedback are merged into one review queue.

JSON
Score breakdown â 2026-05-18-v5
Component Sub-score Weight Contribution
structure_score 85.0 0.15 12.75
security_score 15.0 0.25 3.75
testing_score 100.0 0.20 20.00
documentation_score 91.0 0.15 13.65
practices_score 71.0 0.15 10.65
code_quality 35.7 0.10 3.57
Overall 1.00 64.4
Severity distribution — click a segment to filter
Active filters: excluding tests × Reset all
Corpus Intelligence Cross-corpus context (cohort percentile, top patterns, fix plan) is shown only on repositories you own. Sign up and connect your repo to view it.
Scan summary Strongest dependencies (90), testing (85); weakest security (34), code quality (52). Most common pattern: http-not-https.

Showing 62 of 97 actionable findings. 229 raw detector signals were grouped into reader-sized issues. Click TP / FP to vote on a finding's accuracy — votes adjust the confidence weighting and improve detection across the platform.

critical Security checks quality Quality conf 1.00 ✓ Repobility [MINED030] Python Pickle Loads: pickle.loads() can execute arbitrary code via __reduce__.
Review and fix per the pattern semantics. See CWE-502 / for context.
benchmark/kvbench/runtime/etcd_rt.py:178
critical Security checks quality Quality conf 1.00 [SEC081] Python: pickle.loads / marshal.loads on untrusted data: pickle.load(s) and marshal.load(s) execute arbitrary code on untrusted input. Ported from dlint DUO103 / DUO120 (BSD-3).
Use json, msgpack, or protobuf for untrusted data. If pickle is required, sign the payload with HMAC.
benchmark/kvbench/runtime/etcd_rt.py:178
critical Security checks security secrets conf 0.95 6 occurrences Detected a Generic API Key, potentially exposing access to various services and sensitive operations.
Gitleaks detected a committed secret or credential pattern.
3 files, 6 locations
src/plugins/azure_blob/README.md:127, 134, 143 (3 hits)
benchmark/nixlbench/README.md:861, 878 (2 hits)
.gitlab/test_azure.sh:27
critical Security checks software dependencies conf 0.90 ✓ Repobility GHA script injection via github.event.pull_request.head.ref in run-step
Multi-line `run: |` block interpolates ${{ github.event.pull_request.head.ref }} into shell. PR title/body/branch/comment fields are attacker-controllable.
.github/workflows/build_validation.yml:40
high Security checks security Crypto conf 1.00 3 occurrences [SEC113] SSH host-key verification disabled (MITM): Accepting any SSH host key on first connect lets an active MITM impersonate the server. Common in `paramiko.AutoAddPolicy()`.
Python: load `~/.ssh/known_hosts` and use `paramiko.RejectPolicy()`. Go: implement a `ssh.HostKeyCallback` that compares against a known fingerprint. Java JSch: load known_hosts via `jsch.setKnownHosts(...)`.
3 files, 3 locations
.ci/scripts/run_slurm_allocation.sh:181
.ci/scripts/run_tests_slurm.sh:129
.ci/scripts/stop_slurm_allocation.sh:115
high Security checks quality Quality conf 1.00 ✓ Repobility 12 occurrences `self._configure_gds` used but never assigned in __init__
Method `configure_segment_type` of class `NIXLBench` reads `self._configure_gds`, but no assignment to it exists in __init__ (and no class-level fallback). This raises AttributeError the first time the method runs against an instance.
2 files, 12 locations
benchmark/kvbench/commands/nixlbench.py:261, 263, 265, 267, 445, 485 (6 hits)
benchmark/kvbench/runtime/etcd_rt.py:110, 147, 186, 193, 205, 209 (6 hits)
high Security checks software dependencies conf 0.88 bincode: RUSTSEC-2025-0141
Bincode is unmaintained
Cargo.lock
high Security checks software dependencies conf 0.88 bytes: RUSTSEC-2026-0007
Integer overflow in `BytesMut::reserve`
examples/rust/Cargo.lock
high Security checks cicd CI/CD security conf 0.92 8 occurrences Dockerfile copies the entire context without .dockerignore
COPY . or ADD . sends the full build context to Docker. Without .dockerignore this can include secrets, git history, and local artifacts.
6 files, 8 locations
benchmark/nixlbench/contrib/Dockerfile:84, 251, 252 (3 hits)
.ci/dockerfiles/Dockerfile.gpu-test:27
contrib/Dockerfile:258
contrib/Dockerfile.manylinux:321
contrib/Dockerfile.sglang:20
contrib/Dockerfile.vllm:20
CI/CD securitycontainers
high Security checks software dependencies conf 0.90 ✓ Repobility Dockerfile FROM `gcr.io/google.com/cloudsdktool/cloud-sdk:alpine` not pinned by digest
`FROM gcr.io/google.com/cloudsdktool/cloud-sdk:alpine` resolves the tag at build time. The registry CAN re-push a different image for the same tag, so every build is potentially different. Production images should pin to `image@sha256:...` for reproducibility + supply-chain integrity.
.ci/dockerfiles/Dockerfile.infinia-libs:23
high Security checks cicd CI/CD security conf 0.92 Dockerfile pipes a remote script into a shell
Piping downloaded code directly into a shell bypasses checksum verification and makes builds dependent on mutable remote content.
benchmark/nixlbench/contrib/Dockerfile:73 CI/CD securitycontainers
high Security checks cicd CI/CD security conf 0.92 Dockerfile pipes a remote script into a shell
Piping downloaded code directly into a shell bypasses checksum verification and makes builds dependent on mutable remote content.
.ci/dockerfiles/Dockerfile.build_helper:33 CI/CD securitycontainers
low Security checks cicd CI/CD security conf 0.90 ✓ Repobility 17 occurrences GitHub Action is tag-pinned rather than SHA-pinned
Action `actions/checkout` pinned to mutable ref `@v4` uses a mutable tag or branch. Pin external actions to a reviewed full commit SHA when the workflow is security-sensitive.
9 files, 17 locations
.github/workflows/external_contributor.yaml:30, 48 (4 hits)
.github/workflows/aws_efa_validation.yml:23 (2 hits)
.github/workflows/blossom-ci.yml:46 (2 hits)
.github/workflows/build_validation.yml:16 (2 hits)
.github/workflows/pr-size-check.yml:10 (2 hits)
.github/workflows/pre-commit.yml:9, 15 (2 hits)
.github/workflows/clang-format.yml:9
.github/workflows/copyright-checks.yml:18
CI/CD securitySupply chainGitHub Actions
medium Security checks cicd CI/CD security conf 0.90 ✓ Repobility GitHub Action is tag-pinned rather than SHA-pinned
Action `pre-commit/action` pinned to mutable ref `@v3.0.0` uses a mutable tag or branch. Pin external actions to a reviewed full commit SHA when the workflow is security-sensitive.
.github/workflows/pre-commit.yml:16 CI/CD securitySupply chainGitHub Actions
high Security checks cicd CI/CD security conf 0.90 ✓ Repobility GitHub Action is tag-pinned rather than SHA-pinned
Action `NVIDIA/blossom-action` pinned to mutable ref `@main` uses a mutable tag or branch. Pin external actions to a reviewed full commit SHA when the workflow is security-sensitive.
.github/workflows/blossom-ci.yml:53 CI/CD securitySupply chainGitHub Actions
high Security checks software dependencies conf 0.90 ✓ Repobility 6 occurrences pre-commit hook `https://github.com/pre-commit/mirrors-mypy` pinned to mutable rev `v1.14.1`
`.pre-commit-config.yaml` references `https://github.com/pre-commit/mirrors-mypy` at `rev: v1.14.1`. If `{rev}` is a branch or version tag, the repo owner can push new code there and `pre-commit install --install-hooks` will fetch it on every developer's machine.
lines 16, 22, 28, 34, 41, 48
.pre-commit-config.yaml:16, 22, 28, 34, 41, 48 (6 hits)
high Security checks software dependencies conf 0.88 tracing-subscriber: RUSTSEC-2025-0055
Logging user input may result in poisoning logs with ANSI escape sequences
examples/rust/Cargo.lock
high Security checks software dependencies conf 0.90 ✓ Repobility Workflow container/services image `ghcr.io/ai-dynamo/dynamo/helm-tester:0.1.1` unpinned
`container/services image: ghcr.io/ai-dynamo/dynamo/helm-tester:0.1.1` without `@sha256:...` pulls a mutable tag at workflow-run time. Treat workflow container references with the same supply-chain discipline as Dockerfile FROM lines.
.github/workflows/copyright-checks.yml:10
high System graph hardware Supply chain conf 1.00 Dockerfile pipes a remote installer into a shell
Executing downloaded code during image build gives the remote endpoint build-time code execution. Prefer pinned packages or verify downloaded installers by checksum/signature.
benchmark/nixlbench/contrib/Dockerfile:73 containersRemote installer
high System graph cicd CI/CD security conf 1.00 GitHub Action tracks a moving branch
NVIDIA/blossom-action@main can move without a code change in this repo. Pin third-party actions to a reviewed 40-character commit SHA.
.github/workflows/blossom-ci.yml:53 CI/CD securitySupply chainGithub actions
low Security checks security Deserialization conf 1.00 [SEC007] Unsafe Deserialization: Unsafe deserialization can execute arbitrary code.
Use yaml.safe_load() instead of yaml.load(). Avoid pickle for untrusted data.
benchmark/kvbench/runtime/etcd_rt.py:178
low Security checks quality Error handling conf 0.55 ✓ Repobility Broad exception handler needs review
This handler catches Exception/BaseException. It is actionable when it swallows errors without logging, re-raising, or returning a structured error. Handlers that intentionally convert exceptions into typed error results should not be treated as high risk.
benchmark/kvbench/main.py:192 Error handlingquality
medium Security checks cicd CI/CD security conf 0.90 Docker build context has no .dockerignore
Without .dockerignore, build context can include source history, local env files, dependencies, and generated artifacts.
.dockerignore CI/CD securitycontainers
high Security checks cicd CI/CD security conf 0.82 7 occurrences Docker final stage has no non-root USER
Docker images run as root unless the image or Dockerfile switches to a non-root user.
7 files, 7 locations
.ci/dockerfiles/Dockerfile.gpu-test:11
.ci/dockerfiles/Dockerfile.infinia-libs:24
benchmark/nixlbench/contrib/Dockerfile:109
contrib/Dockerfile:20
contrib/Dockerfile.manylinux:19
contrib/Dockerfile.sglang:17
contrib/Dockerfile.vllm:17
CI/CD securitycontainers
medium Security checks cicd CI/CD security conf 0.90 Dockerfile installs dependencies after copying the full source tree
When dependency installation comes after COPY ., any source change invalidates the dependency layer and makes Docker rebuild much more slowly.
contrib/Dockerfile:291 CI/CD securitycontainers
medium Security checks cicd CI/CD security conf 0.86 2 occurrences Dockerfile separates apt update from install
Splitting apt update and install across layers can reuse stale package indexes and make builds less reliable.
lines 29, 64
benchmark/nixlbench/contrib/Dockerfile:29, 64 (2 hits)
CI/CD securitycontainers
medium Security checks quality Quality conf 1.00 ✓ Repobility Mutable default argument in `create_backend` (dict)
`def create_backend(... = []/{}/set())` — Python's default value is constructed ONCE at function definition time and shared across all calls. Mutating it in one call mutates it for every future call too.
src/api/python/_api.py:382
medium Security checks quality Quality conf 1.00 ✓ Repobility 13 occurrences Mutable default argument in `execute_transfer` (list)
`def execute_transfer(... = []/{}/set())` — Python's default value is constructed ONCE at function definition time and shared across all calls. Mutating it in one call mutates it for every future call too.
2 files, 13 locations
src/api/python/_api.py:150, 402, 425, 466, 495, 552, 599, 716, +4 more (12 hits)
examples/python/remote_storage_example/nixl_p2p_storage_example.py:31
high Security checks software dependencies conf 0.70 Remote install command pipes network code directly to a shell
Agent helper projects often publish one-line installers. `curl | sh` style commands are convenient, but they bypass review unless the script is pinned, signed, or checksum-verified.
src/bindings/rust/README.md:10
high Security checks software dependencies conf 0.70 Remote install command pipes network code directly to a shell
Agent helper projects often publish one-line installers. `curl | sh` style commands are convenient, but they bypass review unless the script is pinned, signed, or checksum-verified.
README.md:226
medium Security checks software dependencies conf 0.90 ✓ Repobility 10 occurrences requirements.txt: `black` has no version pin
Unpinned pip requirement means every fresh install may resolve a different version. Newer releases can introduce malicious code (typosquats, account compromises). Reproducible installs need exact pins.
lines 16, 17, 18, 19, 20, 21, 22, 23, +2 more
requirements.txt:16, 17, 18, 19, 20, 21, 22, 23, +2 more (10 hits)
medium System graph hardware Supply chain conf 1.00 Docker base image uses a mutable or implicit tag: ucx_${UCX}_image
Container tags can be retagged upstream. Pin production base images to a reviewed digest (`image@sha256:...`) when reproducibility and supply-chain integrity matter.
benchmark/nixlbench/contrib/Dockerfile:109 containersPinned dependencies
medium System graph hardware Security conf 1.00 Dockerfile runs as root: benchmark/nixlbench/contrib/Dockerfile
No non-root USER set. Containers running as root expand the blast radius of any vulnerability inside the image.
Container
medium System graph hardware Security conf 1.00 Dockerfile runs as root: contrib/Dockerfile
No non-root USER set. Containers running as root expand the blast radius of any vulnerability inside the image.
Container
medium System graph quality Integrity conf 1.00 Network/subprocess call without timeout or try/except — benchmark/kvbench/commands/nixlbench.py:490
`subprocess.run(...)` here lacks both a `timeout=` arg and an enclosing try/except. This is exactly the class of bug that took down our git-clone earlier (HTTP/2 stream cancel surfaced as a fatal). Add a `timeout=` and wrap in try/except, or use a wrapper that retries.
runtime safetyRobustness
medium System graph security Coverage conf 1.00 No auth library detected
The scanner did not find any standard auth library (JWT, OAuth, NextAuth, Auth0, etc.). Either auth lives in custom code, in a separate service, or is missing.
auth
low Security checks cicd CI/CD security conf 0.72 5 occurrences Dockerfile installs recommended OS packages
Installing recommended packages often pulls in unnecessary runtime surface area.
4 files, 5 locations
benchmark/nixlbench/contrib/Dockerfile:195, 205 (2 hits)
.ci/dockerfiles/Dockerfile.base:36
.ci/dockerfiles/Dockerfile.build_helper:27
contrib/Dockerfile:43
CI/CD securitycontainers
high Security checks cicd CI/CD security conf 0.72 13 occurrences Dockerfile keeps pip download cache
Pip's package cache increases image size and can preserve unnecessary artifacts.
4 files, 13 locations
benchmark/nixlbench/contrib/Dockerfile:243, 245, 249, 267, 290 (5 hits)
contrib/Dockerfile:250, 252, 255, 306 (4 hits)
contrib/Dockerfile.manylinux:290, 292, 295 (3 hits)
contrib/Dockerfile.vllm:22
CI/CD securitycontainers
low Security checks cicd CI/CD security conf 0.74 7 occurrences Dockerfile leaves apt package indexes in the image layer
Package indexes increase image size and can expose stale metadata in the final image layer.
2 files, 7 locations
benchmark/nixlbench/contrib/Dockerfile:53, 86, 195, 205 (4 hits)
contrib/Dockerfile:43, 74, 263 (3 hits)
CI/CD securitycontainers
low Security checks quality Quality conf 0.60 25 occurrences Duplicated implementation block across source files
Duplicate implementation blocks are maintenance debt. Keep them visible, but they are not a high-severity defect unless the duplicated logic is security-sensitive or drifting.
12 files, 19 locations
test/unit/plugins/object/nixl_object_test.cpp:2, 11, 166 (3 hits)
src/plugins/libfabric/libfabric_backend.h:185, 226 (2 hits)
src/plugins/uccl/uccl_backend.h:7, 14 (2 hits)
src/plugins/ucx/ucx_backend.h:44, 58 (2 hits)
test/unit/plugins/infinia/infinia_nixl_test.cpp:90, 213 (2 hits)
test/unit/plugins/posix/nixl_posix_test.cpp:56, 117 (2 hits)
src/bindings/rust/src/descriptors/xfer.rs:72
src/bindings/rust/src/utils/string_list.rs:55
duplicationquality
low System graph hardware Coverage conf 1.00 Containers defined but no K8s/orchestration manifest found
Repo has Dockerfiles/compose but no Kubernetes/Nomad manifests. If the target deployment is K8s, the manifests may live in a separate ops repo.
Deployment
low System graph software Dead code candidate conf 1.00 File has no detected symbols: benchmark/nixlbench/src/runtime/etcd/setup.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: contrib/tomlutil.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: examples/python/nixl_gds_example.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: examples/python/query_mem_example.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: src/bindings/python/nixl-meta/nixl/_api.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: src/bindings/python/nixl-meta/nixl/logging.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph software Dead code candidate conf 1.00 File has no detected symbols: test/python/desc_perf.py
Source file with no class/function declarations — possible config, dead code, or scratch file.
low System graph quality Integrity conf 1.00 5 occurrences Near-duplicate function bodies in 2 places
Functions with the same first-5-line body hash: benchmark/kvbench/test/inference_workload_matgen.py:kv_cache_size, benchmark/kvbench/test/inference_workload_matgen.py:kv_cache_size This is *the* AI-coder failure mode (4× more duplication in vibe-coded repos — see https://jw.hn/ai-code-hygiene). Co…
5 occurrences
repo-level (5 hits)
duplicatesduplication
low System graph quality Integrity conf 1.00 Old/deprecated-named symbol `zero_copy` in examples/device/ep/nixl_ep/buffer.py:447
Names with suffixes like `_old`, `_v1`, `_deprecated` usually indicate replaced-but-not-removed code (typical AI-coder leftover). Confirm and delete, or rename if it's the active version.
old markerDead code
low System graph quality Integrity conf 1.00 Old/deprecated-named symbol `zero_copy` in examples/device/ep/tests/elastic/elastic.py:324
Names with suffixes like `_old`, `_v1`, `_deprecated` usually indicate replaced-but-not-removed code (typical AI-coder leftover). Confirm and delete, or rename if it's the active version.
old markerDead code
low System graph software Dead code conf 1.00 Possibly dead Python function: ctp_args
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
benchmark/kvbench/commands/args.py:285
low System graph software Dead code conf 1.00 Possibly dead Python function: destroy_dist
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
benchmark/kvbench/runtime/rt_base.py:61
low System graph software Dead code conf 1.00 Possibly dead Python function: destroy_dist
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
benchmark/kvbench/runtime/etcd_rt.py:90
low System graph software Dead code conf 1.00 Possibly dead Python function: estimate_xfer_cost
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
src/api/python/_api.py:526
low System graph software Dead code conf 1.00 Possibly dead Python function: invalidate_target_metadata
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
examples/python/partial_md_example.py:55
low System graph software Dead code conf 1.00 Possibly dead Python function: make_connection
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
src/api/python/_api.py:466
low System graph software Dead code conf 1.00 Possibly dead Python function: query_xfer_backend
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
src/api/python/_api.py:679
low System graph software Dead code conf 1.00 Possibly dead Python function: set_tcp_store_group
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
examples/device/ep/nixl_ep/buffer.py:787
low System graph software Dead code conf 1.00 Possibly dead Python function: to_yaml
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
benchmark/kvbench/models/model_config.py:206
low System graph software Dead code conf 1.00 Possibly dead Python function: update_notifs
No callers detected by AST scan in this repo. Could be exported for external callers or a framework handler.
src/api/python/_api.py:730
low System graph quality Complexity conf 1.00 Very large file: src/bindings/rust/tests/tests.rs (2272 lines)
Files with >800 lines often hide complexity hotspots and discourage tests.
For AI agents: Voting guide (TP/FP) MCP manifest Stdio wrapper SARIF Integrate Findings queue Vote TP/FP on findings to calibrate the engine.
For AI agents + API integrations
Email me when this repo regresses
Free. We re-scan periodically; new criticals → your inbox. No signup required for the scan itself.
API access

This page is publicly accessible at: https://repobility.com/scan/53d6037f-ca55-47aa-82ef-91636bc47761/

To check status programmatically (no auth required):

curl -s https://repobility.com/api/v1/public/scan/53d6037f-ca55-47aa-82ef-91636bc47761/

Important — please don't re-submit the same URL repeatedly. The submission endpoint is idempotent: re-submitting the same git URL returns this same scan_token, not a new one. To re-scan this repo, sign up free and use the dashboard.