Compare commits

...
Author SHA1 Message Date
KaifAhmad1 bf292ccbbc fix(docker): address terrascan findings, drop apt-get upgrade
Two terrascan/GHAS findings on the previous commit:
- AC_DOCKER_0052 (no apt-get upgrade in Dockerfiles): dropped it. It also
  wasn't fixing anything - Debian's openssl fix for CVE-2026-14456 is still
  in trixie-proposed-updates, not reachable via a normal upgrade. Pin both
  base images by digest instead (matches #1329's approach) so the docker
  Dependabot ecosystem bumps them once Debian ships a rebuilt image with
  the fix, and document why the QUIC DoS isn't reachable here regardless
  (HTTP-only via uvicorn).
- AC_DOCKER_0010 (pin pip package versions): setuptools was `>=78.1.1`;
  pinned to the exact 84.0.0 already used by pyproject.toml/requirements-ci.txt.

Also fixes two build breaks this introduces on its own: requirements-ci.txt
wasn't in .dockerignore's allowlist or container-scan.yml's path trigger,
so the COPY in the prior commit would have failed the image build outright.
2026-08-31 14:06:28 +05:30
KaifAhmad1 64d942503b fix(docker): extract requirements-ci.txt pins with Python instead of sed
The sed expression to strip requirements-ci.txt's line-continuation
backslash (`[\]$`) is valid POSIX/GNU sed - verified it exits 0 and
strips correctly - but it's easy to misread as broken (a bot reviewer
flagged it as an unterminated bracket expression), and the seemingly
more obvious `\$`/` \$` forms silently fail to match at all rather
than erroring. Swap to a small `re.findall` extraction so there's no
backslash-escaping judgment call left for a reader (bot or human) to
second-guess.
2026-08-31 14:06:02 +05:30
KaifAhmad1 471a420b10 fix(docker): resolve Trivy-flagged CVEs in the built image
Container Security Scan flagged five HIGH-severity findings against
semantica:scan:
- setuptools 70.3.0 (CVE-2025-47273, path traversal) - the base image's
  bundled copy, never touched by our own build. Upgraded explicitly.
- msgpack 1.1.2 (GHSA-6v7p-g79w-8964, OOB read/crash) - `pip install
  ".[explorer]"` re-resolved deps from scratch instead of reusing the
  audited, hash-pinned requirements-ci.txt (which already pins
  msgpack==1.2.1), so it landed on an unpatched transitive version. Now
  installs against a constraints file derived from requirements-ci.txt.
- openssl / libssl3t64 / openssl-provider-legacy (CVE-2026-14456, QUIC
  server DoS) - the Debian fix is still in trixie-proposed-updates, not
  yet promoted to trixie-security, so it can't be pulled via apt today.
  Added an apt upgrade step so the next image rebuild picks it up
  automatically once Debian ships it; documented why this image isn't
  actually exposed to it in the meantime (HTTP-only via uvicorn, no QUIC
  listener).
2026-08-31 14:06:02 +05:30
Mohd KaifandSameer6305 a4aa71ad87 fix(ci): unblock py3.9 install matrix and raise Scorecard pinning/signing (#1329)
* fix(ci): unblock py3.9 install matrix and raise Scorecard pinning/signing

pip install semantica failed on Python 3.9 across all three OSes because
spacy had no upper bound, so pip resolved spacy 3.8.16 whose thinc>=8.3.12
requirement has no cp39 wheels and no working sdist build path. Cap
spacy/thinc for python_version < '3.10' to the last wheel-compatible pair.

Also addresses the two OpenSSF Scorecard findings that were actually
fixable in code:
- Pinned-Dependencies: Dockerfile base images (node:26-alpine,
  python:3.13-slim) were unpinned by digest; pin both, and pin five
  previously-unversioned pip install calls in CI (build, safety, bandit,
  semgrep, jq, pip-audit).
- Signed-Releases: attest-build-provenance only publishes to the GH
  attestations API, which Scorecard doesn't inspect. Sign dist/* with
  Sigstore and attach the .sigstore.json bundles as release assets.

* fix(ci): correct Sigstore artifact inputs

---------

Co-authored-by: Sameer6305 <sskadam6305@gmail.com>
2026-08-31 14:04:28 +05:30
8 changed files with 60 additions and 11 deletions
+1
View File
@@ -6,6 +6,7 @@
!README.md
!LICENSE
!MANIFEST.in
!requirements-ci.txt
!semantica/
!semantica/**
!integrations/
+1 -1
View File
@@ -72,7 +72,7 @@ jobs:
diff \
<(grep -E '^[a-zA-Z0-9._-]+==' requirements-ci.txt | sed 's/ \\$//') \
<(grep -E '^[a-zA-Z0-9._-]+==' /tmp/requirements-ci-check.txt)
- run: pip install build
- run: pip install build==1.6.0
# wheel is build-time only (not in requirements-ci.txt) — install the
# same pinned version [build-system] declares so --no-isolation works.
- run: pip install wheel==0.48.0
+1
View File
@@ -12,6 +12,7 @@ on:
- 'README.md'
- 'LICENSE'
- 'MANIFEST.in'
- 'requirements-ci.txt'
- 'semantica/**'
- 'integrations/**'
- 'explorer/**'
+17 -3
View File
@@ -16,7 +16,7 @@ jobs:
cancel-in-progress: false
permissions:
contents: write # for the GitHub Release
id-token: write # for PyPI Trusted Publishing (OIDC) and attestation signing
id-token: write # for PyPI Trusted Publishing (OIDC), attestation signing, and Sigstore
attestations: write # for SLSA build provenance
# If you add another job to this workflow, give it its own explicit
# `permissions:` block rather than relying on the workflow-level default
@@ -40,7 +40,7 @@ jobs:
# build runs against the same versions CI tests against.
- name: Install pinned build dependencies
run: pip install -r requirements-ci.txt
- run: pip install build
- run: pip install build==1.6.0
# wheel is build-time only (not in requirements-ci.txt) — install the
# same pinned version [build-system] declares so --no-isolation works.
- run: pip install wheel==0.48.0
@@ -71,7 +71,21 @@ jobs:
uses: actions/attest-build-provenance@4d101475d8b20a2381f78447822ac1eab6504dd8 # v4
with:
subject-path: 'dist/*'
# attest-build-provenance publishes to the GH attestations API only, which
# OpenSSF Scorecard's Signed-Releases check does not inspect - it looks for
# signature files attached as release assets. Sign here too so
# `dist/*.sigstore.json` bundles ship alongside the wheel/sdist on the
# GitHub Release itself.
- name: Sign artifacts with Sigstore
uses: sigstore/gh-action-sigstore-python@790bc6befb9d733738f18d8f895854b453640ec9 # v3.5.0
with:
inputs: |
dist/*.whl
dist/*.tar.gz
- uses: softprops/action-gh-release@efb35369e0ad2afab669f228072c1b0d510eae64 # v3.0.3
with:
files: dist/*
files: |
dist/*.whl
dist/*.tar.gz
dist/*.sigstore.json
- uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+1 -1
View File
@@ -52,7 +52,7 @@ jobs:
# Tooling AFTER the pinned set: installing safety/bandit/semgrep/jq
# first lets the pinned requirements overwrite their transitive deps
# (e.g. rich), which breaks the safety CLI at runtime.
pip install safety bandit semgrep jq
pip install safety==3.8.1 bandit==1.9.4 semgrep==1.175.0 jq==1.12.0
- name: Run Safety Check (Package Vulnerabilities)
run: |
+1 -1
View File
@@ -37,6 +37,6 @@ jobs:
# pyproject.toml changes under review. The schedule/workflow_dispatch
# runs stay non-blocking until a full pass over pre-existing findings
# across the whole [all] tree has been done.
- run: pip install pip-audit
- run: pip install pip-audit==2.10.1
- run: pip-audit -r requirements-ci.txt
continue-on-error: ${{ github.event_name != 'pull_request' }}
+30 -4
View File
@@ -1,5 +1,5 @@
# syntax=docker/dockerfile:1
FROM node:26-alpine AS frontend-builder
FROM node:26-alpine@sha256:2d984a15c9b54fd0aeb608b8e0d0d83529eb34d2966db27a1fb4f1edc3d298a3 AS frontend-builder
WORKDIR /app
COPY explorer/package*.json ./explorer/
@@ -9,7 +9,18 @@ RUN npm ci
COPY explorer/ ./
RUN mkdir -p /app/semantica && npm run build
FROM python:3.13-slim AS runtime
# CVE-2026-14456 (OpenSSL QUIC-server DoS, flagged against this base image's
# openssl/libssl3t64/openssl-provider-legacy): the Debian fix
# (3.5.7-1~deb13u2) is only in trixie-proposed-updates as of this writing,
# not yet promoted to trixie-security, so there's no package to pin here
# today. Deliberately NOT running `apt-get upgrade` to chase it - that
# breaks build reproducibility (terrascan AC_DOCKER_0052) and still
# wouldn't reach a proposed-updates-only package. Once Debian ships the fix
# and rebuilds this tag, the docker Dependabot ecosystem in
# .github/dependabot.yml opens a PR bumping the digest pin above. Also: this
# image only serves plain HTTP via uvicorn and never opens a QUIC listener,
# so the bug isn't reachable here regardless.
FROM python:3.13-slim@sha256:7ce4b6dfe35e55397b7cda544f8a13f191b7ae28dc5aad71fe664dbc9bc2623f AS runtime
ENV PYTHONDONTWRITEBYTECODE=1 \
PYTHONUNBUFFERED=1 \
@@ -22,12 +33,27 @@ WORKDIR /app
RUN groupadd --system semantica \
&& useradd --system --gid semantica --home-dir /app --shell /usr/sbin/nologin semantica
COPY pyproject.toml README.md LICENSE MANIFEST.in ./
COPY pyproject.toml README.md LICENSE MANIFEST.in requirements-ci.txt ./
COPY semantica/ ./semantica/
COPY integrations/ ./integrations/
COPY --from=frontend-builder /app/semantica/static ./semantica/static
RUN pip install --no-cache-dir ".[explorer]" \
# The base image ships an outdated setuptools (CVE-2025-47273); upgrade it
# explicitly since nothing in our own dependency tree otherwise pulls a
# newer copy. Pinned to the exact version requirements-ci.txt/pyproject.toml
# already build against, rather than a floor, per terrascan AC_DOCKER_0010.
# requirements-ci.txt itself carries the audited, CVE-checked pins for every
# transitive dependency (see security-scan.yml / security.yml) - feed them
# in as an unhashed constraints file (pip's hash-checking mode rejects the
# unhashable local source directory this installs) so the image lands on
# the same patched versions CI verified, e.g. msgpack>=1.2.1, rather than
# letting pip freely re-resolve and pick up an unpatched transitive version.
# (Extracted with Python's re module rather than sed/grep so there's no
# line-continuation-backslash stripping to get subtly wrong.)
RUN pip install --no-cache-dir "setuptools==84.0.0" \
&& python -c "import re, pathlib; pins = re.findall(r'^([A-Za-z0-9._-]+==\S+)', pathlib.Path('requirements-ci.txt').read_text(), re.M); pathlib.Path('/tmp/constraints.txt').write_text('\n'.join(pins))" \
&& pip install --no-cache-dir -c /tmp/constraints.txt ".[explorer]" \
&& rm -f /tmp/constraints.txt requirements-ci.txt \
&& chown -R semantica:semantica /app
USER semantica
+8 -1
View File
@@ -49,7 +49,14 @@ dependencies = [
"scipy>=1.13.1",
"scikit-learn>=1.7.2",
"umap-learn>=0.5.12",
"spacy>=3.4.0",
# thinc (spacy's core dep) dropped Python 3.9 wheels at 8.3.10, and later
# spacy patch releases (3.8.8+) require thinc>=8.3.9-only-on-3.10+ ranges,
# which forces a source build that fails outright on 3.9 (see Install
# Matrix run history). Capping both keeps 3.9 on the last wheel-compatible
# pair; 3.10+ is left unconstrained to always get the latest spacy/thinc.
"spacy>=3.4.0,<3.8.8; python_version < '3.10'",
"spacy>=3.4.0; python_version >= '3.10'",
"thinc<8.3.5; python_version < '3.10'",
"transformers>=4.20.0",
"torch>=1.13.1",
"sentence-transformers>=2.2.0",