[Ian Hellen (ianhelle)]

Summary

Prepare the 3.1.1 security hotfix without deprecating the Mordor driver.

• Escape provider-controlled TI browser headings, nested/scalar details, raw responses and reference text. Only absolute HTTP/HTTPS references become clickable links.
• Replace the two remaining MITRE pickle caches with table-oriented JSON. Legacy pickle caches are never read or migrated; invalid JSON caches are logged and refreshed. Preserve text-column dtypes, indices and null values, and support connections with disk caching disabled.
• Keep python-publish.yml, but disable both jobs with unconditional false guards. Separate build and publish, remove static PyPI token use, use OIDC with a pypi environment, retain full action SHA pins, reduce permissions and restrict future execution to non-prerelease v* releases.
• Replace the repository-owned MITRE test fixtures with JSON and update injected cache preparation in both GitHub Actions and Azure Pipelines, including refreshing metadata timestamps.
• Update directly related documentation and bump the patch version from 3.1.0 to 3.1.1.
• Pin Pylint to 4.0.9, the last passing main CI version, in dev dependencies and pre-commit so new lint checks are adopted deliberately rather than introduced by an unconstrained CI upgrade.

Investigation

The workflow's mutable-tag reports were already addressed by previous SHA-pinning changes. Mordor commit 2bca6952 replaced the metadata pickle cache, but left the techniques/tactics pickle caches. Remote metadata and normal dataset downloads use known HTTPS sources; the remaining deserialization risk was an attacker replacing locally writable pickle caches, not an established remote-code-execution route through the upstream datasets.

Validation

• Focused TI browser, Mordor driver and notebook widget tests: 76 passed, 3 skipped. Existing network/notebook skip behavior is preserved.
• JSON cache round-trip, stale/corrupt/schema-invalid cache refresh, legacy-pickle avoidance, uncached connections and HTML/link injection regression coverage.
• Executed both CI cache preparation scripts with network access blocked: each loaded 695 techniques, 14 tactics and 96 datasets from the injected fixtures.
• YAML, JSON and whitespace pre-commit hooks passed for changed files; scoped Ruff lint/format checks, pylint and pydocstyle passed.
• The exact whole-package CI Pylint command passed with Pylint 4.0.9 / Astroid 4.0.4. The previous failed PR run had installed Pylint 4.1.3 / Astroid 4.3.4.
• Full mypy passed for all 273 source files using --python-version 3.13, matching the installed interpreter. The default 3.10 target cannot parse the installed NumPy stubs; no repository type-check configuration was changed.
• Verified packaging and runtime version 3.1.1.

Publishing configuration

This PR disables execution in YAML; it does not change GitHub repository settings or configure PyPI. Before deliberately re-enabling publishing, configure the protected pypi environment with required approval and release-tag restrictions, and register the matching PyPI trusted publisher. The legacy PYPI_API_TOKEN is no longer referenced by this workflow.