.github/workflows/sign.yml, called by the GitHub mirror's ci.yml on every v* tag after the index, relay index and release exist:
Signs the server and inbuxa Relay multi-arch indexes by digest with cosign, keyless (the workflow's GitHub identity, logged in Sigstore's public log), in the registry and on the GHCR copy, then verifies as a user would.
SBOMs: syft, SPDX, one per architecture; attested against each architecture's image with GitHub's attest-sbom, and attached to the Gitea release.
Provenance: GitHub's attest-build-provenance against each index, only from the run that built it.
By hand, from main with an existing tag: signs and attests SBOMs for a version published before this (or whose signing failed); no build attestation, since that run built nothing. Plan after merge: run it for v27.1-rc.33 and verify end to end before rc.34 depends on it.
Supply chain: cosign v3.1.4 and syft v1.55.0 (both a month old) downloaded at pinned versions and checked against pinned SHA-256 sums; actions pinned to commit SHAs; only GitHub's own actions and the docker/* ones. Attestations are kept in GitHub's store keyed by digest, so nothing depends on what Gitea's registry stores beside an image.
sign is in the report, so the tag's Gitea check fails if signing does. The forum announcement still goes out when the release is published (announce.yml), before signing.
Not tested end to end: keyless signing only runs on GitHub. Tested here: workflow YAML parses, the shell steps pass bash -n, the digest and per-architecture lookups against rc.33 on both registries, the tag pattern (rc, release, patch and 2026.x dated tags). Fork checks clean. The docs page on checking signatures follows once a signed image exists.
Closes #430 (K-S7). Keyless, as decided.
**`.github/workflows/sign.yml`**, called by the GitHub mirror's `ci.yml` on every `v*` tag after the index, relay index and release exist:
- **Signs** the server and inbuxa Relay multi-arch indexes by digest with cosign, keyless (the workflow's GitHub identity, logged in Sigstore's public log), in the registry and on the GHCR copy, then verifies as a user would.
- **SBOMs**: syft, SPDX, one per architecture; attested against each architecture's image with GitHub's `attest-sbom`, and attached to the Gitea release.
- **Provenance**: GitHub's `attest-build-provenance` against each index, only from the run that built it.
**By hand**, from main with an existing tag: signs and attests SBOMs for a version published before this (or whose signing failed); no build attestation, since that run built nothing. Plan after merge: run it for `v27.1-rc.33` and verify end to end before rc.34 depends on it.
**Checking an image:**
```
cosign verify <image>@<digest> \
--certificate-oidc-issuer https://token.actions.githubusercontent.com \
--certificate-identity-regexp '^https://github.com/inbuxa/inbuxa-server/\.github/workflows/sign\.yml@refs/(tags/v.+|heads/main)$'
gh attestation verify oci://<image>@<digest> --owner inbuxa
```
**Supply chain**: cosign v3.1.4 and syft v1.55.0 (both a month old) downloaded at pinned versions and checked against pinned SHA-256 sums; actions pinned to commit SHAs; only GitHub's own actions and the docker/* ones. Attestations are kept in GitHub's store keyed by digest, so nothing depends on what Gitea's registry stores beside an image.
`sign` is in the report, so the tag's Gitea check fails if signing does. The forum announcement still goes out when the release is published (announce.yml), before signing.
**Not tested end to end**: keyless signing only runs on GitHub. Tested here: workflow YAML parses, the shell steps pass `bash -n`, the digest and per-architecture lookups against rc.33 on both registries, the tag pattern (rc, release, patch and 2026.x dated tags). Fork checks clean. The docs page on checking signatures follows once a signed image exists.
.github/workflows/sign.yml, called by ci.yml on every v* tag once the
images, their index and the release exist:
- cosign signs the server's and inbuxa Relay's multi-arch index by
digest, keyless (this workflow's GitHub identity, recorded in Sigstore's
public log), in the registry and on the GHCR copy, then verifies the
signature as a user would.
- syft writes an SPDX SBOM per architecture; GitHub's attest-sbom attests
each against that architecture's image, and they are attached to the
Gitea release.
- attest-build-provenance attests the build against each index, only
from the run that built it.
Run by hand from main with an existing release's tag, it signs and
attests the SBOMs of a version published before signing (or whose
signing failed), and attests no build: that run built nothing.
The attestations live in GitHub's store, keyed by digest, so nothing
depends on what the registry keeps beside an image. cosign v3.1.4 and
syft v1.55.0 are downloaded at pinned versions and checked against
pinned SHA-256 sums; the actions are pinned to commit SHAs.
ci.yml: the sign job is part of the report, so the tag's check on Gitea
fails when signing does. A GHCR failure only leaves the copy unsigned.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Closes #430 (K-S7). Keyless, as decided.
.github/workflows/sign.yml, called by the GitHub mirror'sci.ymlon everyv*tag after the index, relay index and release exist:attest-sbom, and attached to the Gitea release.attest-build-provenanceagainst each index, only from the run that built it.By hand, from main with an existing tag: signs and attests SBOMs for a version published before this (or whose signing failed); no build attestation, since that run built nothing. Plan after merge: run it for
v27.1-rc.33and verify end to end before rc.34 depends on it.Checking an image:
Supply chain: cosign v3.1.4 and syft v1.55.0 (both a month old) downloaded at pinned versions and checked against pinned SHA-256 sums; actions pinned to commit SHAs; only GitHub's own actions and the docker/* ones. Attestations are kept in GitHub's store keyed by digest, so nothing depends on what Gitea's registry stores beside an image.
signis in the report, so the tag's Gitea check fails if signing does. The forum announcement still goes out when the release is published (announce.yml), before signing.Not tested end to end: keyless signing only runs on GitHub. Tested here: workflow YAML parses, the shell steps pass
bash -n, the digest and per-architecture lookups against rc.33 on both registries, the tag pattern (rc, release, patch and 2026.x dated tags). Fork checks clean. The docs page on checking signatures follows once a signed image exists.