Grades are a hygiene check, not a security guarantee.

Claude Code, Skill

Gpu Perf Tune

GPU performance engineering for agents and MCP clients, with an inference-focused skills package and reusable cluster utilities.

Compatibility: Not stated.

Grade C

One Medium finding. Read the report before installing.

Graded by PlugVetRubric v2

Not a guarantee. A hygiene check of the source at the commit shown, not a security audit.

NETWORK_HOST_FROM_INPUT

Why this grade

  • NETWORK_HOST_FROM_INPUT Medium

    A file takes a URL or endpoint as input and has no fixed host list.

    plugins/profile-and-optimize/server/tools/stage-model-parallel.py, line 66 with socket.create_connection((u.hostname, port), timeout=timeout):

Any High, or two or more Medium findings, give D. One Medium gives C. Low findings or a security Note give B. Each code names a finding behind the letter.

Grades are a hygiene check of the source at the commit shown, on the date checked, not a security audit or a guarantee. The free directory grades are automated. Before release, a PlugVet engineer reviews every paid report and any grade we hold back, including all F grades, against the source code and the evidence we quote.

What was checked

Plain notes from the hygiene checks.

  • Permissions

    Note

    requests shell access. The manifest names this permission.

    PERMISSION_SHELL_DECLARED

    plugins/profile-and-optimize/hooks/provenance-audit.py, line 26

    result = subprocess.run(

    plugins/profile-and-optimize/server/tools/test_hooks.py, line 24

    process = subprocess.run(

    plugins/profile-and-optimize/server/tools/test_hooks.py, line 123

    subprocess.run(["git", "init", "-q", str(repo)], check=True)

    Show more

    plugins/profile-and-optimize/server/tools/test_hooks.py, line 124

    subprocess.run(["git", "-C", str(repo), "add", str(source)], check=True)

    plugins/profile-and-optimize/server/tools/test_runtime_scripts.py, line 106

    subprocess.run([BASH, str(script)], env=env, check=True)

    plugins/profile-and-optimize/server/tools/test_runtime_scripts.py, line 107

    subprocess.run([BASH, str(script)], env=env, check=True)

    plugins/profile-and-optimize/server/tools/test_runtime_scripts.py, line 111

    subprocess.run([BASH, str(script)], env=env, check=True)

    plugins/profile-and-optimize/server/tools/test_runtime_scripts.py, line 145

    result = subprocess.run(

  • Install and update

    Pass

    No install-time script found, and dependencies are pinned or absent.

  • Network

    Serious

    accepts a host from input, with no fixed list in this file.

    NETWORK_HOST_FROM_INPUT

    plugins/profile-and-optimize/server/tools/stage-model-parallel.py, line 66

    with socket.create_connection((u.hostname, port), timeout=timeout):

    plugins/profile-and-optimize/server/tools/stage-model-parallel.py, line 177

    def fetch(key, size, out):

    plugins/profile-and-optimize/server/tools/perf_tune_report/orchestrator/production_steps.py, line 21

    warmup ``kubectl exec curl /v1/models`` retry up to 60s

    Show more

    plugins/profile-and-optimize/server/tools/perf_tune_report/runners/settle.py, line 49

    " urllib.request.urlopen(r, timeout=600).read()\n"

    plugins/profile-and-optimize/server/tools/perf_tune_report/runners/spec_metrics.py, line 54

    f"m = urllib.request.urlopen('{base}/metrics', timeout=30).read().decode()\n"

    plugins/profile-and-optimize/server/tools/profiling/hang_detector/gpusd_scraper.py, line 23

    ``curl http://<node>:<port>/metrics > snapshot.json`` and replay it

    plugins/profile-and-optimize/skills/inference-decode-step-budget/SKILL.md, line 81

    start["curl -XPOST localhost:8000/start_profile"]

    plugins/profile-and-optimize/skills/inference-decode-step-budget/SKILL.md, line 83

    stop["curl -XPOST localhost:8000/stop_profile"]

    plugins/profile-and-optimize/skills/zymtrace-anchored-query/SKILL.md, line 117

    if ! curl -fsS -o /dev/null -m 2 'http://localhost:9123/ping'; then

    scripts/check-doc-links.sh, line 16

    # when curl is missing or --no-network is passed (cluster login nodes

    scripts/check-doc-links.sh, line 249

    # GitHub orgs whose repos return 404 to unauthenticated curl HEAD requests even

    scripts/check-doc-links.sh, line 257

    """Single-URL curl HEAD invocation. Designed to be thread-pool-friendly."""

    scripts/check-doc-links.sh, line 279

    # Internal repo / blob; unauth curl would return 404. Trust that the

    scripts/check-doc-links.sh, line 284

    return True, "skipped (curl missing)"

  • Secrets

    Concern

    reads a credential or environment file.

    SECRETS_ENV_READ SECRETS_PATH_DOTENV

    plugins/profile-and-optimize/server/tools/stage-model-parallel.py, line 55

    AK = os.environ.get("PERFLAKE_LAKE_S3_ACCESS_KEY") or os.environ.get("AWS_ACCESS_KEY_ID")

    plugins/profile-and-optimize/server/tools/stage-model-parallel.py, line 56

    SK = os.environ.get("PERFLAKE_LAKE_S3_SECRET_KEY") or os.environ.get("AWS_SECRET_ACCESS_KEY")

    plugins/profile-and-optimize/server/tools/perf_tune_report/lake_writer.py, line 2122

    perflake ``.env``.

    Show more

    plugins/profile-and-optimize/server/tools/perf_tune_report/lake_writer.py, line 2258

    "./.env; set +a`), or pass "

    plugins/profile-and-optimize/server/tools/profile_and_optimize_mcp/scripts/configure_clients.py, line 126

    f"[mcp_servers.{SERVER_NAME}.env]",

    plugins/profile-and-optimize/skills/inference-aa-workload/assets/repro_artificialanalysis.py, line 312

    default=os.environ.get("API_KEY", os.environ.get("WANDB_INFERENCE_API_KEY", "")),

    plugins/profile-and-optimize/skills/inference-aa-workload/assets/repro_artificialanalysis.py, line 322

    p.add_argument("--tokenizer", default=os.environ.get("TOKENIZER", "") or None)

    .gitignore, line 9

    dist/

    .gitignore, line 10

    .env

    plugins/profile-and-optimize/server/docs/secrets.md, line 9

    Keep a non-secret `.env.example`-style template for local variable names, and

    plugins/profile-and-optimize/server/docs/secrets.md, line 10

    keep the real `.env` untracked and scoped to the current workstation or

    plugins/profile-and-optimize/skills/inference-kernel-ncu-profile/SKILL.md, line 197

    env (`kubectl get deploy <name> -o jsonpath='{...env...}'`) before assuming a

  • Agent instructions

    Pass

    No instruction to ignore earlier instructions or to hide a step.

  • Obfuscation

    Note

    the file cap was reached before every file was read.

    OBFUSCATION_FILE_CAP_REACHED

    scripts/extract-release-notes.py, line 1

    #!/usr/bin/env python3

  • Licence

    Pass

    A licence file is present (LICENSE).

  • Maintainer activity

    Pass

    The latest commit is within the last 365 days.

Note

Hygiene check of commit 25d74d031732. requests shell access. The manifest names this permission. accepts a host from input, with no fixed list in this file. reads a credential or environment file. the file cap was reached before every file was read.

Date and source

Date checked:

View source

github.com

Commit graded: 25d74d031732e2f76d30a43e44c3434b588f06b3

This link goes to the original host. PlugVet does not host or check what you download from there, and installing it is your own decision and risk.

Read how this letter is chosen