Supply-chain security

What gets signed, which credentials exist, and what the checks actually prove.

What the Checks Prove#

Tests, formatting, pedantic Clippy and RustSec gate every standard release, and MSRV joins them when configured. None of that makes a binary trustworthy. It makes it reviewable: every artifact traces back to a run, a commit and a set of checks somebody can inspect.

Corprus Crucible signs the binary with keyless Cosign before it is packaged, and the verification bundle travels inside the archive. Verify the extracted binary against that bundle and the expected workflow identity. The archive itself is not signed, so do not tell anyone it is. VirusTotal then scans the archive and records the links. A clean scan means no scanner complained. It does not mean the binary is benign.

Credentials#

  • GITHUB_TOKEN: built in, limited by the caller's and the job's permissions.
  • id-token: write: the OIDC identity keyless signing and Pages deployment rely on.
  • VT_API_KEY: needed for every non-PR binary release.
  • CARGO_REGISTRY_TOKEN: crates.io, only when publishing is on.
  • AUR_SSH_PRIVATE_KEY: AUR publishing.
  • NEXUS_API_KEY and NEXUS_GROUP_IDS: Nexus uploads, always as a pair.
  • DW_BOT_PAT: opening update issues in dependent repositories.
  • Discord webhooks: notifications only. A missing webhook skips quietly, by design.

The input reference shows which contract declares what. No secret value belongs anywhere under war-room/ or in a plan; this whole site is public.

Workflow Scripts#

A run: step receives caller-supplied text, like a notification title, through env:, never spliced into the script. ${{ inputs.title }} pasted into shell source means a title containing $(...) gets executed. discord.yml shows the pattern.

When Something Goes Wrong#

Keep the run URL, commit or tag, caller ref, job, target and artifact identity. If some channels already published before the failure, find out which before retrying. Then the incident desk and the recovery rules.