Skip to content

fix(ci): harden workflows, fuzz harness, and dev scripts for v1.9.2 - #173

Draft
skupriienko wants to merge 1 commit into
masterfrom
wave-1-tooling-hygiene
Draft

skupriienko wants to merge 1 commit into
masterfrom
wave-1-tooling-hygiene

Conversation

@skupriienko

@skupriienko skupriienko commented Nov 15, 2026 •

Copy link
Copy Markdown
Collaborator

Stack (v1.9.2 Hardening Rollout)

Wave PR Base <- Head Scope
-> Wave 1 (This PR) #173 master <- wave-1-tooling-hygiene CRIT-03, CRIT-04, CRIT-05, HIGH-06
Wave 2 #174 wave-1-tooling-hygiene <- wave-2-transport-redaction CRIT-01, CRIT-02, HIGH-01 (lock), HIGH-02, HIGH-03, HIGH-04 (path), OPT-09
Wave 3 #175 wave-2-transport-redaction <- wave-3-boot-slots-pooling OPT-01..OPT-07, OPT-10, OPT-11, HIGH-01 (warn_unclosed)
Wave 4 #176 wave-3-boot-slots-pooling <- wave-4-builders-semver HIGH-04 (bool/stream), HIGH-05, HIGH-07, OPT-08
Wave 5 #177 wave-4-builders-semver <- wave-5-typing-stateful-tests MOD-02, MOD-04

Merge in order: #173 -> #174 -> #175 -> #176 -> #177.


Summary

Wave 1 of 5 for the v1.9.2 hardening rollout. Hardens GitHub Actions workflows (zizmor-clean, OIDC Trusted Publishing), activates all 8 guardrails fuzz targets (0..7), fixes manage.sh and Makefile hygiene (zero *.txt deletes, strict pip-audit exit codes, set -Eeuo pipefail), and aligns errors.py and README.md docstrings with exception-raising HTTP error semantics.


Actions

  • Trusted Publishing & CI Workflows (.github/workflows/, CRIT-05):
    • Updated .github/workflows/publish.yml to enforce PyPI/TestPyPI OIDC Trusted Publishing (id-token: write), restore IS_STABLE gating, and remove static token bypasses.
    • Updated .github/workflows/commit_checks.yaml, codeql.yml, issue-triage.yml, and security.yml to drop live MJ_APIKEY_* env injection from the offline CI matrix, add mypy --strict, pyright, slotscheck, tests/test_boot.py, and tests/test_perf.py gates, fix codeql.yml target branch (master) and codeql-action/init step, and add issue-triage.yml null guards.
  • Guardrails Fuzz Targets 0..7 (tests/fuzz/fuzz_guardrails.py, tests/unit/test_fuzz_targets.py, .clusterfuzzlite/, CRIT-03):
    • Expanded ConsumeIntInRange selector to (0, 7) so targets 0..7 execute cleanly instead of leaving targets 5..7 unreachable.
    • Hardened .clusterfuzzlite/Dockerfile and .clusterfuzzlite/build.sh and added offline unit dispatcher reachability tests in tests/unit/test_fuzz_targets.py.
  • Developer Script & Makefile Hygiene (manage.sh, Makefile, environment-dev.yaml, pyproject.toml, CRIT-04):
    • Applied set -Eeuo pipefail, removed broad *.txt deletion in clean rules, stopped swallowing exit codes in pip-audit and fuzz_all, and wired mypy --strict, pyright, and slotscheck into make analyze.
  • Exception Documentation Alignment (mailjet_rest/errors.py, README.md, samples/*, HIGH-06):
    • Updated stale docstrings and examples claiming 4xx responses return requests.Response without raising so docs match the exception-raising client behavior (ValidationError, MailjetAuthError, DoesNotExistError, ApiRateLimitError, MailjetApiError).

Verification & Testing

  • Verified offline checks across the v1.9.2 stack: ruff check ., ruff format --check ., mypy --strict mailjet_rest, pyright mailjet_rest, slotscheck -m mailjet_rest, bandit -c pyproject.toml -r mailjet_rest, zizmor .github/workflows/ (0 findings), pip-audit --skip-editable, and pytest -n auto -W error tests/unit tests/property tests/regression.
  • Standalone atheris fuzz execution runs in ClusterFuzzLite CI (atheris C extension is unavailable locally on Python 3.15); target 0..7 dispatcher coverage is verified offline in tests/unit/test_fuzz_targets.py.
  • Live Mailjet API network tests (pytest -m network, samples/smoke_readme_runner.py) were skipped locally due to absent MJ_APIKEY_PUBLIC / MJ_APIKEY_PRIVATE credentials.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant