sign vulnerability attestation before saving it to the disk

[developer-guy]

We (w/@Dentrax) thought there was a slight time window that the attacker could gain access to the system and changes the content of the vulnerability attestation before you sign and upload it to the registry. A similar problem has been addressed in the Syft project for SBOM attestations¹. So, we should sign the vulnerability attestation before saving it to the disk.

If you like the idea, we are willing to do that, and please share your thoughts, thanks.
cc: @knqyf263 @itaysk

Footnotes

1. https://anchore.com/sbom/creating-sbom-attestations-using-syft-and-sigstore/ ↩

[knqyf263]

What if we pass attestations via pipe?

$ trivy image -f cosin-vuln <IMAGE> | COSIGN_EXPERIMENTAL=1 cosign attest --type vuln --predicate - <IMAGE>

Then, all security scanners don't have to re-implement cosign functionalities.

[developer-guy]

Quoted from @znewman01:

FWIW you can do this today in bash or zsh (though not sh) with process substitution:

COSIGN_EXPERIMENTAL=1 cosign attest \
    --type vuln \
    --predicate <(trivy image -f cosin-vuln <IMAGE>) \
    <IMAGE>

[znewman01]

But also note @developer-guy's PR to accept - to read from STDIN: sigstore/cosign#2215

That's a better long-term solution 😄

[knqyf263]

Looks like @developer-guy'S PR got merged. Thanks for your help.

[itaysk]

We should probably update the docs still