Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/sigstore-scaffolding-e2e.yaml:
- Around line 133-137: The workflow step consumes outputs
steps.versions.outputs.mkcert and
steps.versions.outputs.mkcert_sha256_linux_amd64 that are not defined; add those
outputs to the version-loading action and ensure the source settings include
corresponding keys (e.g., mkcert and mkcert_sha256_linux_amd64) so they
populate, or alternatively hardcode/derive MKCERT_VERSION and MKCERT_SHA256 in
the workflow; update the load-versions action (declare outputs with those names)
and the settings data source (add mkcert and mkcert_sha256_linux_amd64 entries)
so the Install mkcert step receives non-empty values.

In
`@tests/chainsaw/signing/bundle-attestation-private-sigstore/chainsaw-test.yaml`:
- Around line 128-135: Replace brittle grep string checks in the test that
inspect "${BUNDLE}" with semantic JSON assertions using jq: instead of grep for
"verificationMaterial", "certificate", and "rawBytes" validate that
.verificationMaterial.certificate.rawBytes exists and is a non-empty string,
assert that .verificationMaterial.publicKey is absent for the keyless path, and
for checksum results assert .checksumsPassed is the boolean true (not a
whitespace-sensitive string match). Locate usages of the BUNDLE variable and the
checks around verificationMaterial.certificate.rawBytes and checksumsPassed in
the chainsaw-test.yaml and change them to jq-based queries that fail the test if
the field is missing, empty, or not the expected type/value.

In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/run.sh`:
- Around line 158-167: The wait_url function can hang because curl calls lack
per-request timeouts; update the curl invocations inside wait_url to include
connection and total time limits (e.g., --connect-timeout 5 and --max-time 10)
so each poll attempt is bounded, and use the same timeout flags when capturing
the final failing output (the last="$(curl ... || true)") so the error message
reflects a timed-out attempt rather than a stuck TCP connect; modify the curl
usages within wait_url (the loop check and the final diagnostic curl)
accordingly.

In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/tlsproxy/main.go`:
- Around line 100-107: The server is currently bound to all interfaces via Addr:
":" + listen in the srv := &http.Server {...} block; change Addr to bind to
loopback only (e.g. "127.0.0.1:"+listen or use net.JoinHostPort("127.0.0.1",
listen)) so the TLS proxy only listens on localhost. Update the same pattern for
the other server instance(s) referenced in the file (the similar Addr
assignments around lines 112-113) to ensure both proxies are loopback-only;
leave all other fields (Handler, TLSConfig, ReadHeaderTimeout) unchanged.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/sigstore-scaffolding-e2e.yaml:
- Around line 133-145: The inline shell step "Install mkcert" (and similarly the
steps "Build aicr binary", "Locate binary and detect attestation", and "Run
private-sigstore e2e") must either be moved into a Layer-2 composite action or
be annotated with a one-line rationale above the run: block explaining why the
inline implementation is intentional (e.g., "inline by design / single-use: this
step contains short, non-reusable orchestration specific to this workflow and
extracting a composite would add indirection without reuse"). Update each
retained inline run to include that exact brief justification comment or
refactor the logic into a composite action to comply with the three-layer
composite actions architecture.

In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/README.md`:
- Around line 3-52: The README uses the acronyms "OIDC" and "SA" without
defining them; update README.md to define these on first use (replace "OIDC"
with "OpenID Connect (OIDC)" and "SA" with "ServiceAccount (SA)" where they
first appear) and/or add a short "Terminology" section near the top defining
OIDC and SA (and optionally Fulcio and Rekor) so readers see full terms before
acronyms are reused.

---

Duplicate comments:
In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/run.sh`:
- Around line 156-167: The wait_url function uses curl without per-attempt
timeouts, so a single hung request can exceed the intended ~90s polling window;
update the curl invocations inside wait_url (both the success-check curl and the
final diagnostic curl stored in last) to include sensible timeouts (e.g.,
--connect-timeout and --max-time) so each attempt is bounded, and ensure the
diagnostic curl uses the same timeouts when capturing the last error; adjust the
timeout values to keep total polling duration ≈90s given 45 attempts with 2s
sleeps.

In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/tlsproxy/main.go`:
- Around line 100-113: The server is currently bound to all interfaces via Addr
":" + listen in the srv http.Server, which exposes the test TLS proxy
externally; change the binding to loopback-only (e.g. "127.0.0.1:"+listen or
"localhost:"+listen) in the srv initialization where Addr is set (and update the
log message using the same listen binding if needed) so the proxy only listens
on the local host; update references to listen/targetURL in that block (srv,
listen, targetURL, and the slog.Info call) to reflect the loopback binding.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In
`@tests/chainsaw/signing/bundle-attestation-private-sigstore/chainsaw-test.yaml`:
- Around line 183-185: Update the stale comment that references "grep" to
reflect the current `yq` JSON evaluation: replace the phrase "The grep above
fails (non-zero) on a checksum mismatch" with something like "The yq assertion
above fails (non-zero) on a checksum mismatch" so the comment next to the
checksum check (the comment describing why the checksum check is independent of
verify's exit code and the expected private-Rekor attestation failure)
accurately describes the `yq` assertion being used.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/actions/load-versions/action.yml:
- Around line 94-105: The composite action duplicates outputs named mkcert,
mkcert_sha256_linux_amd64, and mkcert_sha256_linux_arm64 (and their
corresponding echo assignments), which breaks action parsing; remove the
duplicated output entries and the duplicate echo/set-output assignments so each
output (scaffold_chart, mkcert, mkcert_sha256_linux_amd64,
mkcert_sha256_linux_arm64) is defined only once and populated only once from
steps.versions.outputs. Locate the repeated definitions of the mkcert* outputs
in the outputs block and the repeated echo lines (the second set around the
195-206/204-206 region) and delete the redundant entries to restore a single
canonical output assignment for each mkcert* symbol.

In `@tests/chainsaw/signing/bundle-attestation-private-sigstore/run.sh`:
- Around line 53-55: The script currently assumes goreleaser is present when
AICR_BIN is unset; update the build fallback to match the stated prerequisites
by checking for goreleaser and falling back to go build when goreleaser is
missing (or alternatively add a preflight check that ensures goreleaser is
installed before using it). Concretely, in the AICR_BIN unset branch (the block
that invokes goreleaser), replace the hard dependency with a conditional: if
command -v goreleaser >/dev/null then run goreleaser, else run go build (or set
an error with a clear message prompting installation), and also add a
corresponding preflight check for goreleaser where environment/tools are
verified so the script behavior and the “goreleaser or go” prereq list stay
aligned.