CI(security): catch Lightning, Shai-Hulud, npm hijack, design-flaw CVEs

Recent supply-chain incidents that scan_packages would have missed:
  - PyTorch Lightning 2.6.x: payload in _runtime/router_runtime.js
    (14.8 MB), persistence via .claude/settings.json SessionStart
    and .vscode/tasks.json folderOpen
  - npm chalk/debug + Shai-Hulud: hex-var obfuscation, window.ethereum
    Web3 hijack, .github/workflows/shai-hulud.yml repo takeover,
    trufflehog credential exfil
  - elementary-data 0.23.3: token harvesters with embedded gh{p,o,s}_
    and AKIA regexes
  - litellm 1.82.7: also covered by existing patterns, but anyone on
    `>=` got it during the 40-min exposure window
  - langchain-core CVE-2025-68664 / n8n CVE-2025-68668 / marimo
    CVE-2026-39987: first-party design flaws, not malicious-author

scan_packages.py:
  - Six new regexes: RE_DEV_TOOL_HIJACK, RE_TOKEN_REGEX,
    RE_JS_OBFUSCATION, RE_WEB3_HIJACK, RE_WORKFLOW_INJECT,
    RE_SHELL_DROPPER.
  - Three new checkers: check_js_file, check_shell_file,
    check_workflow_file. scan_archive now routes .js/.mjs/.cjs/.ts
    to the JS checker, .sh/.bash to the shell checker, and
    .github/workflows/*.yml to the workflow checker.
  - JS checker fires CRITICAL on hex-var obfuscation OR Web3 hijack
    OR (token regex + network) OR workflow-injection signature; HIGH
    on a >100 KB JS bundle inside a Python wheel (the Lightning tell).
  - Smoke-tested: every new pattern matches its canonical positive
    and rejects four legitimate-looking false-positive baits.

security-audit.yml:
  - OSV-Scanner step: cross-ecosystem advisory check (PyPI + npm
    + cargo) from one binary. OSV's feed is a superset of GitHub-
    Advisory; catches CVEs that haven't propagated yet (e.g.
    langchain-core was on OSV before GitHub Advisory).
  - Semgrep step: p/supply-chain + p/python + p/javascript +
    p/security-audit packs catch first-party logic bugs (CVEs 7/9/10
    above) that pattern scanning never sees.
  - Lockfile pin verifier: warns on every non-`==` spec in
    requirements/*.txt. Currently surfaces 104 unpinned specs as
    informational baseline; tighten to blocking once the baseline
    is curated.

All new steps continue-on-error initially; they surface findings to
the workflow summary + advisory-audit-logs artifact.
This commit is contained in:
Daniel Han 2026-05-06 23:54:04 +00:00
commit 1696a15cf8
2 changed files with 332 additions and 0 deletions

View file

@ -261,6 +261,126 @@ jobs:
echo '```'
} >> "$GITHUB_STEP_SUMMARY"
# ─────────────────────────────────────────────────────────────
# OSV-Scanner: cross-ecosystem advisory DB (PyPI + npm + cargo)
# ─────────────────────────────────────────────────────────────
- name: OSV-Scanner (PyPI + npm + cargo, cross-ecosystem advisories)
# OSV's advisory feed is a superset of GitHub-Advisory + RustSec
# + npm advisories; running it alongside the per-ecosystem audit
# tools catches CVEs that haven't propagated to the per-ecosystem
# DBs yet (e.g. langchain-core CVE-2025-68664 was on OSV before
# GitHub Advisory). Single binary, one transitive resolver, all
# three lockfile types in one pass. Non-blocking until baselines
# close.
continue-on-error: true
run: |
set +e
curl -fsSL -o /tmp/osv-scanner.tar.gz \
https://github.com/google/osv-scanner/releases/download/v2.0.2/osv-scanner_linux_amd64.tar.gz
tar -xzf /tmp/osv-scanner.tar.gz -C /tmp osv-scanner
/tmp/osv-scanner --version
/tmp/osv-scanner scan source \
--lockfile=studio/frontend/package-lock.json \
--lockfile=studio/src-tauri/Cargo.lock \
--lockfile=requirements.txt:audit-reqs/unsloth-deps.txt \
--lockfile=requirements.txt:audit-reqs/studio.txt \
--lockfile=requirements.txt:audit-reqs/no-torch-runtime.txt \
--lockfile=requirements.txt:audit-reqs/overrides.txt \
--lockfile=requirements.txt:audit-reqs/extras.txt \
--lockfile=requirements.txt:audit-reqs/extras-no-deps.txt \
--format=table 2>&1 | tee logs-osv-scanner.txt
{
echo "## OSV-Scanner (cross-ecosystem)"
echo
echo '```'
tail -200 logs-osv-scanner.txt
echo '```'
} >> "$GITHUB_STEP_SUMMARY"
# ─────────────────────────────────────────────────────────────
# Semgrep: design-flaw detection (catches what regex-pattern
# scanning of malicious authors cannot — first-party logic bugs
# like langchain-core CVE-2025-68664 dumps/dumpd injection,
# n8n CVE-2025-68668 _pyodide.eval_code sandbox escape, marimo
# CVE-2026-39987 unauth WebSocket).
# ─────────────────────────────────────────────────────────────
- name: Semgrep (supply-chain + python rule packs)
continue-on-error: true
run: |
set +e
python -m pip install --quiet 'semgrep>=1.95'
semgrep --version
semgrep scan \
--config p/supply-chain \
--config p/python \
--config p/javascript \
--config p/security-audit \
--severity ERROR --severity WARNING \
--metrics off \
--timeout 120 \
studio/backend unsloth scripts \
2>&1 | tee logs-semgrep.txt
{
echo "## Semgrep (supply-chain + python + javascript rules)"
echo
echo '```'
tail -200 logs-semgrep.txt
echo '```'
} >> "$GITHUB_STEP_SUMMARY"
# ─────────────────────────────────────────────────────────────
# Lockfile pin verifier. The litellm 1.82.7 attack window was
# ~40 minutes; anyone resolving with `>=` got the malicious
# version automatically. Flag every spec in the requirements
# files that does not pin to an exact `==` (or `@` for git
# refs, or `===` for arbitrary equality). Warning-only for now;
# graduate to blocking once the baseline is clean.
# ─────────────────────────────────────────────────────────────
- name: Lockfile pin verifier (Python requirements)
continue-on-error: true
run: |
python <<'PY' | tee logs-pin-verifier.txt
import re
from pathlib import Path
# Specs that look like `pkg==1.2.3` or `pkg @ git+...` or
# bare comments / -r lines are pinned-or-not-applicable.
PINNED = re.compile(r"^\s*[A-Za-z0-9_.\-]+\s*(?:===|==)\s*[^,;]+\s*$")
GIT_OR_URL = re.compile(r"^\s*[A-Za-z0-9_.\-]+\s*@\s*(?:git\+|https?://)")
unpinned = []
for f in sorted(Path("studio/backend/requirements").glob("*.txt")):
for i, raw in enumerate(f.read_text().splitlines(), 1):
line = raw.strip()
if not line or line.startswith("#") or line.startswith("-"):
continue
spec = line.split("#", 1)[0].strip().split(";", 1)[0].strip()
if not spec:
continue
if "git+" in spec or PINNED.match(spec) or GIT_OR_URL.match(spec):
continue
unpinned.append((str(f), i, line))
print(f"::group::Lockfile pin status")
if unpinned:
print(f"WARN: {len(unpinned)} non-`==` specs across requirements/*.txt")
print("(litellm 1.82.7 wave hit anyone on `>=`; tighten when feasible.)")
for f, i, line in unpinned[:80]:
print(f" {f}:{i}: {line}")
if len(unpinned) > 80:
print(f" ... and {len(unpinned) - 80} more")
else:
print("OK: every spec is exact-pinned.")
print("::endgroup::")
PY
{
echo "## Lockfile pin verifier"
echo
echo '```'
cat logs-pin-verifier.txt
echo '```'
} >> "$GITHUB_STEP_SUMMARY"
- uses: actions/upload-artifact@v4
if: always()
with:
@ -270,6 +390,9 @@ jobs:
logs-npm-audit.txt
logs-npm-audit.json
logs-cargo-audit.txt
logs-osv-scanner.txt
logs-semgrep.txt
logs-pin-verifier.txt
audit-reqs/
retention-days: 30

View file

@ -306,6 +306,84 @@ RE_C2_POLLING = re.compile(
re.DOTALL,
)
# Developer-tool persistence hooks. The PyTorch Lightning 2.6.x compromise
# planted SessionStart hooks into Claude Code, VS Code tasks, and Cursor
# settings so the payload re-attached on every editor open. Catches any
# package writing into a known dev-tool config that supports auto-run.
RE_DEV_TOOL_HIJACK = re.compile(
r"\.claude/settings\.json"
r"|\.cursor/.*hooks"
r"|\.vscode/(?:tasks|settings|launch)\.json"
r"|SessionStart|folderOpen|onCommand:.*runTask"
r"|/etc/profile\.d/"
r"|\b\.bashrc\b|\b\.zshrc\b|\b\.profile\b"
r"|\bautomator\b.*\.workflow\b",
)
# Hard-coded credential / API-token regexes embedded in source. Packages
# that ship regexes for OTHER people's secrets are nearly always
# stealers (litellm 1.82.7, elementary-data 0.23.3, Shai-Hulud).
RE_TOKEN_REGEX = re.compile(
r"\bgh[psoru]_[A-Za-z0-9_]{20,}" # GitHub PAT/OAuth/etc.
r"|\bgithub_pat_[A-Za-z0-9_]{20,}"
r"|\bnpm_[A-Za-z0-9]{30,}" # npm token
r"|\bsk-[A-Za-z0-9]{20,}" # OpenAI / Anthropic
r"|\bxox[bpaesr]-" # Slack
r"|\bAIza[0-9A-Za-z_-]{20,}" # Google API key
r"|\bAKIA[0-9A-Z]{16}" # AWS access key id
r"|\bASIA[0-9A-Z]{16}" # AWS STS
r"|\bgithub.com/login/oauth/access_token"
r"|\bglpat-[0-9A-Za-z_-]{20,}", # GitLab PAT
)
# JavaScript-side obfuscation. The npm chalk/debug compromise and the
# Lightning router_runtime.js use the same minifier-style hex-var name
# pattern; a bundle full of `_0x1f2e3d` identifiers is a near-universal
# tell for a malicious npm payload (and very rare in legit minified code
# that ships in PyPI wheels).
RE_JS_OBFUSCATION = re.compile(
r"_0x[a-f0-9]{4,6}\s*=\s*function"
r"|var\s+_0x[a-f0-9]{4,6}\b"
r"|(?:\\x[0-9a-f]{2}){10,}" # \x-escape strings
r"|String\.fromCharCode\s*\(\s*\d+\s*(?:,\s*\d+\s*){10,}\)",
)
# Web3 / wallet-hijack pattern. The Qix npm phish overrode fetch /
# XMLHttpRequest and attached a `window.ethereum` listener that
# Levenshtein-swapped recipient addresses on the way to the network.
RE_WEB3_HIJACK = re.compile(
r"\bwindow\.ethereum\b"
r"|\bweb3\.eth\.\w+\s*\("
r"|XMLHttpRequest\.prototype\.(?:open|send)\s*="
r"|(?:^|\s)fetch\s*=\s*\(?\s*async"
r"|TronWeb|solanaWeb3",
)
# Self-propagating supply-chain worms (Shai-Hulud, ForceMemo) plant
# their own GitHub workflow in every repo they can reach, and lean on
# trufflehog/gitleaks for credential discovery. The combo of any of
# these strings inside a *package payload* is overwhelming evidence of
# repo-takeover intent.
RE_WORKFLOW_INJECT = re.compile(
r"\.github/workflows/[^\"\']*\.ya?ml"
r"|\btrufflehog\b|\bgitleaks\b"
r"|/user/repos\?affiliation=.*owner.*collaborator"
r"|\bshai-hulud\b|EveryBoiWeBuildIsAWormyBoi"
r"|\bgit\s+push\s+--force\b.*--no-verify",
re.IGNORECASE | re.DOTALL,
)
# Shell-side patterns specific to install.sh / postinstall scripts that
# pipe remote code into a shell. `curl ... | sh` and friends are the
# canonical npm postinstall dropper.
RE_SHELL_DROPPER = re.compile(
r"\bcurl\b[^\n|]*\|\s*(?:sh|bash|zsh)\b"
r"|\bwget\b[^\n|]*-O-\s*\|\s*(?:sh|bash|zsh)\b"
r"|\bnpx\b\s+-y\s+[^\s]+@latest\s*\|"
r"|\beval\s+\$\(\s*curl\b"
r"|\bbash\s+<\(\s*curl\b",
)
# ---------------------------------------------------------------------------
# Finding dataclass
@ -868,6 +946,125 @@ def _extract_evidence(content: str, pattern: re.Pattern, max_matches: int = 3) -
return " | ".join(matches) if matches else ""
# ---------------------------------------------------------------------------
# Non-Python checkers
# ---------------------------------------------------------------------------
# Several recent PyPI compromises (PyTorch Lightning 2.6.x, ForceMemo)
# carried the active payload in a bundled .js / .sh / workflow yaml so
# the Python imports looked clean on first glance. These checkers scan
# those file types when they appear inside a Python wheel/sdist.
def check_js_file(content: str, filename: str, package: str) -> list[Finding]:
"""Run JS-side checks. Triggered by .js / .mjs / .cjs / .ts."""
findings = []
# A JS file *inside a Python wheel* that's larger than 100 KB is
# itself anomalous (legit Python packages don't ship hand-written
# JS bundles). Combined with ANY of the other JS heuristics it is
# CRITICAL; standalone it is HIGH.
is_large = len(content) > 100 * 1024
has_obf = bool(RE_JS_OBFUSCATION.search(content))
has_web3 = bool(RE_WEB3_HIJACK.search(content))
has_token_regex = bool(RE_TOKEN_REGEX.search(content))
has_workflow_inj = bool(RE_WORKFLOW_INJECT.search(content))
has_network = bool(RE_NETWORK.search(content))
if has_obf:
sev = CRITICAL if (is_large or has_web3 or has_token_regex) else HIGH
findings.append(Finding(
sev, package, filename,
"JS minifier-style hex-var obfuscation (npm-payload signature)",
_extract_evidence(content, RE_JS_OBFUSCATION),
))
if has_web3:
findings.append(Finding(
CRITICAL, package, filename,
"JS Web3 / wallet hijack (window.ethereum or fetch override)",
_extract_evidence(content, RE_WEB3_HIJACK),
))
if has_token_regex and has_network:
findings.append(Finding(
CRITICAL, package, filename,
"JS embeds credential regexes AND makes network calls (stealer)",
_extract_evidence(content, RE_TOKEN_REGEX),
))
if has_workflow_inj:
findings.append(Finding(
CRITICAL, package, filename,
"JS self-propagation: workflow injection / repo takeover signature",
_extract_evidence(content, RE_WORKFLOW_INJECT),
))
if is_large and not findings:
findings.append(Finding(
HIGH, package, filename,
f"Python wheel ships large ({len(content) // 1024} KB) JS bundle "
"(uncommon; manually review)",
"",
))
return findings
def check_shell_file(content: str, filename: str, package: str) -> list[Finding]:
"""Run shell-side checks. Triggered by .sh / .bash / install scripts."""
findings = []
if RE_SHELL_DROPPER.search(content):
findings.append(Finding(
CRITICAL, package, filename,
"Shell pipes remote code into an interpreter (curl|sh dropper)",
_extract_evidence(content, RE_SHELL_DROPPER),
))
if RE_DEV_TOOL_HIJACK.search(content) and (
RE_NETWORK.search(content) or RE_SUBPROCESS.search(content)
):
findings.append(Finding(
CRITICAL, package, filename,
"Shell installs developer-tool persistence hook (.bashrc / "
"profile.d / vscode tasks) AND has network or exec",
_extract_evidence(content, RE_DEV_TOOL_HIJACK),
))
if RE_TOKEN_REGEX.search(content) and RE_NETWORK.search(content):
findings.append(Finding(
CRITICAL, package, filename,
"Shell embeds credential regexes AND makes network calls",
_extract_evidence(content, RE_TOKEN_REGEX),
))
if RE_WORKFLOW_INJECT.search(content):
findings.append(Finding(
CRITICAL, package, filename,
"Shell self-propagation: workflow injection / repo takeover signature",
_extract_evidence(content, RE_WORKFLOW_INJECT),
))
return findings
def check_workflow_file(content: str, filename: str, package: str) -> list[Finding]:
"""Run GitHub-Actions workflow checks. Triggered by .github/workflows/*.yml."""
findings = []
# A GitHub workflow file inside a *PyPI package* is itself
# suspicious (Shai-Hulud's whole MO is to plant `shai-hulud.yml`
# in every repo it can write to). Anything matching the workflow
# injection signature gets flagged CRITICAL.
if RE_WORKFLOW_INJECT.search(content):
findings.append(Finding(
CRITICAL, package, filename,
"Workflow file inside PyPI package matches self-propagation signature",
_extract_evidence(content, RE_WORKFLOW_INJECT),
))
if RE_TOKEN_REGEX.search(content):
findings.append(Finding(
HIGH, package, filename,
"Workflow file embeds credential regexes (token harvesting?)",
_extract_evidence(content, RE_TOKEN_REGEX),
))
if RE_SHELL_DROPPER.search(content):
findings.append(Finding(
CRITICAL, package, filename,
"Workflow pipes remote code into a shell (curl|sh dropper)",
_extract_evidence(content, RE_SHELL_DROPPER),
))
return findings
# ---------------------------------------------------------------------------
# Archive handling
# ---------------------------------------------------------------------------
@ -916,6 +1113,18 @@ def scan_archive(archive_path: str, package: str) -> list[Finding]:
findings.extend(check_pth_file(content, filename, package))
elif lower.endswith(".py"):
findings.extend(check_py_file(content, filename, package))
elif lower.endswith((".js", ".mjs", ".cjs", ".ts")):
# Lightning 2.6.x hid its real payload in a 14.8 MB
# router_runtime.js inside a Python wheel. Without this
# branch we'd have only seen the small Python loader.
findings.extend(check_js_file(content, filename, package))
elif lower.endswith((".sh", ".bash")):
findings.extend(check_shell_file(content, filename, package))
elif "/.github/workflows/" in lower and lower.endswith((".yml", ".yaml")):
# Shai-Hulud / ForceMemo plant their own GHA workflow.
# A workflow file inside a *PyPI package* is on its own
# already a yellow flag; pattern-match the worm signatures.
findings.extend(check_workflow_file(content, filename, package))
return findings