hMailServer

Most workflow_dispatch inputs are handled safely: sign-release.yml passes inputs.tag through an env var and exits if empty; upstream-watch, sbom and codeql use the same env-var pattern; server-build interpolates only type:choice/boolean inputs that GitHub validates against fixed options. Gap: installer-smoke.yml interpolates the free-form string input release_tag directly into a pwsh run block (gh release download "${{ inputs.release_tag }}", lines 46 and 50), so a crafted value could inject commands on a runner holding github.token. That collaborator input is not sanitized or validated before shell use. See https://github.com/Progressiverobot/hmailserver/blob/master/.github/workflows/installer-smoke.yml

No documented policy for managing the project's own secrets and credentials exists on master: README.md, .github/SECURITY.md, CONTRIBUTING.md, SUPPORT.md, RELEASE.md and ARCHITECTURE.md contain nothing on storing, accessing or rotating project credentials. Practice is good — secret scanning and push protection are enabled, release signing is Sigstore keyless so no long-lived signing key exists, and workflows use only the ephemeral github.token — but the required written policy is missing. PR #40's GOVERNANCE.md inventories credential-bearing 'Critical assets' and who holds them, yet gives no storage/access/rotation guidelines, so merging it alone would not fully meet this. See https://github.com/Progressiverobot/hmailserver/pull/40

No durable project documentation tells users how to verify release assets. On master, README.md, RELEASE.md, Roadmap.md and docs/RegulatoryScope.md describe the signing design but give no verification commands, and the notes of the current Latest release v6.2.21 contain none (checked back through v6.2.19). The only published instructions are in the v6.2.23-alpha1 prerelease notes (2026-08-21): a complete cosign verify-blob command with bundle, identity regexp and OIDC issuer, plus the expected-failure statement. A one-off prerelease note is not documentation a user of the supported release would find; the command needs a stable home such as README or a VERIFYING.md. See https://github.com/Progressiverobot/hmailserver/releases/tag/v6.2.23-alpha1

The expected signer identity — a Sigstore certificate identity matching ^https://github.com/Progressiverobot/hmailserver/ with issuer https://token.actions.githubusercontent.com — is stated only once, in the v6.2.23-alpha1 prerelease notes published 2026-08-21. No repository documentation on master (README.md, SECURITY.md, RELEASE.md, docs/) records the expected identity, and the notes of the current Latest release v6.2.21 do not mention it. Signing is real (every asset gets a keyless cosign bundle, verified in-workflow before upload), but a user has no stable documented statement of which identity to expect. See https://github.com/Progressiverobot/hmailserver/releases/tag/v6.2.23-alpha1

Partially in place but not yet for all released assets. Every release since v6.2.3 (June 2026), including the latest stable v6.2.21 and all current prereleases, ships SPDX and CycloneDX JSON SBOMs alongside the compiled installer; they are generated by Syft in .github/workflows/sbom.yml (which also merges the native OpenSSL/Boost/libpq dependencies Syft cannot see and fails the job if any is missing), and RELEASE.md step 12 makes SBOM attachment a mandatory pre-publication step. However, six earlier releases that remain published (v6.0.0-B1, v6.0.0, v6.1.0, v6.2.0, v6.2.1, v6.2.2) contain only the compiled installer with no SBOM, so not all compiled released software assets are delivered with an SBOM. Remediation: those six releases predate GitHub's immutable-releases setting (immutable:false), so SBOMs can be attached retroactively by dispatching the SBOM workflow with release_tag for each legacy tag, or the legacy releases can be retired; either would make this criterion Met.

Unmet. Master has no branch protection (GitHub API returns 404 "Branch not protected") and no required-review rule; the sole maintainer routinely commits directly to master without any pull request or second reviewer. This is a single-maintainer project (494 of 498 commits by one person) with no second human available to approve changes. The project acknowledges this openly: Roadmap.md notes "Expect Branch-Protection and Code-Review to fail by design on a single-maintainer repository." A tag ruleset protects release tags, but that does not review commits. See https://github.com/Progressiverobot/hmailserver/blob/master/Roadmap.md

Unmet. No VEX document or feed exists — not in the repository (repo-wide grep finds no VEX content) and not among release assets (v6.2.21 ships installer, SBOMs and signatures only). Dependabot currently shows 0 open and 0 dismissed alerts, but it only covers the .NET/Actions ecosystems; the vendored native binaries (e.g. 7-Zip 19.00 from 2019, flagged in ThirdPartyBinaries.md as overdue a refresh) are assessed narratively in ThirdPartyBinaries.md and SECURITY.md's out-of-scope section, not in machine-readable VEX augmenting the SBOMs. See https://github.com/Progressiverobot/hmailserver/blob/master/hmailserver/docs/ThirdPartyBinaries.md

Unmet. No documented policy defines a remediation threshold for SCA findings covering both vulnerabilities and licenses. What exists: the dependency-review workflow blocks PR-introduced dependencies with known high/critical CVEs (fail-on-severity: high, rationale documented in the workflow header) and Dependabot files grouped update PRs — but there is no license-finding threshold anywhere, no documented timeline for remediating vulnerabilities discovered in already-shipped dependencies, and SECURITY.md's 90-day target covers externally reported vulnerabilities, not SCA findings. See https://github.com/Progressiverobot/hmailserver/blob/master/.github/workflows/dependency-review.yml

Unmet. There is no documented policy requiring SCA violations to be resolved before a release, and no pre-release status check verifies it. RELEASE.md is a detailed 13-step release checklist (freeze, adversarial review, full regression suite, SBOM attachment, signing) but contains no step to check or clear Dependabot alerts or other SCA findings before shipping. The dependency-review gate runs only on pull requests, not on the release process, and the sole maintainer's direct pushes bypass it. See https://github.com/Progressiverobot/hmailserver/blob/master/RELEASE.md

Unmet. Automated SCA exists but does not cover all changes and cannot block. The dependency-review workflow evaluates every pull request against the GitHub Advisory Database and fails on high/critical CVEs (fail-on-severity: high) — but master has no branch protection (API returns 404), so no status check is required and a failing check does not block a merge; and the sole maintainer's direct pushes to master, which are the normal way changes land here, are never evaluated at all since the workflow triggers only on pull_request. See https://github.com/Progressiverobot/hmailserver/blob/master/.github/workflows/dependency-review.yml

Unmet. No documented policy defines a threshold for remediation of SAST findings. What exists: CodeQL runs on both languages (C# security-and-quality on every push/PR; C++ security-extended weekly/on-demand), .github/codeql/codeql-config.yml records each excluded rule with measured counts and reasoning rather than silent suppression, and Roadmap.md acknowledges a "Static-analysis backlog" whose remainder "needs triage rather than blanket suppression" — an honest status note, not a policy stating which finding severities must be fixed and in what timeframe. See https://github.com/Progressiverobot/hmailserver/blob/master/Roadmap.md

Unmet. SAST is not a blocking gate on all changes. CodeQL C# runs on every push/PR but publishes alerts to code scanning without any required status check — master has no branch protection (API 404), so nothing is blocked on violations. CodeQL C++ — covering the internet-facing protocol parsers — runs only weekly and on manual dispatch; codeql.yml itself states plainly that "an alert introduced by a pull request is not seen until the next weekly run on master." Direct pushes to master by the sole maintainer, the normal change path, face no SAST gate at all. See https://github.com/Progressiverobot/hmailserver/blob/master/.github/workflows/codeql.yml