Skip to content

chore(toolbox): bump pinned tools to clear the drift gate - #91

Merged
garymike merged 1 commit into
mainfrom
automated/tool-bumps
Oct 30, 2026
Merged

garymike merged 1 commit into
mainfrom
automated/tool-bumps

Conversation

@garymike

Copy link
Copy Markdown
Owner

Opens the branch that tool-bump.yml pushed on 2026-10-26 (commit 71c5de0, server-signed), so the weekly Tool drift gate can go green again. It has been red since 2026-09-28; the last four Mondays' runs all failed on scripts/check-tool-updates.sh --fail-on-drift.

Applied by tool-bump.yml and validated by building every image from these pins before the branch was written.

Bumped

Tool From To Files
trufflehog 3.99.3 3.99.7 toolbox/tools.lock x1, toolbox/base/Dockerfile
snyk-agent-scan 0.6.8 0.6.9 toolbox/tools.lock x1, toolbox/mcp-review/Dockerfile
semgrep 1.181.0 1.182.1 toolbox/tools.lock x2, toolbox/skill-audit/Dockerfile, toolbox/sast/Dockerfile
checkov 3.3.29 3.3.30 toolbox/tools.lock x1, toolbox/iac/Dockerfile

After merging

Build toolbox images will publish new digests for every image, and its pins job will go red until the five reusable-workflow digest pins move to what it published (the same follow-up as #85 and #89). The 2026-10-26 scheduled rebuild already left those pins behind; merging this first avoids moving them twice.

Opened by the weekly github-security-audit run.

Applied by tool-bump.yml and validated by building every image
from these pins before the branch was written.

### Bumped

| Tool | From | To | Files |
|---|---|---|---|
| trufflehog | `3.99.3` | `3.99.7` | toolbox/tools.lock x1, toolbox/base/Dockerfile |
| snyk-agent-scan | `0.6.8` | `0.6.9` | toolbox/tools.lock x1, toolbox/mcp-review/Dockerfile |
| semgrep | `1.181.0` | `1.182.1` | toolbox/tools.lock x2, toolbox/skill-audit/Dockerfile, toolbox/sast/Dockerfile |
| checkov | `3.3.29` | `3.3.30` | toolbox/tools.lock x1, toolbox/iac/Dockerfile |
@garymike
garymike merged commit 4ff8c1d into main Oct 30, 2026
9 checks passed
@garymike
garymike deleted the automated/tool-bumps branch October 30, 2026 18:31
garymike added a commit that referenced this pull request Oct 30, 2026
…blished (#95)

The second half of #91 and #93, and the first pins that scripts/check-image-pins.py can
actually verify: 02c62e4 is the first build that stamps each image with a fingerprint of its
build inputs, so before this every pin reported UNVERIFIABLE.

Resolved from each image's `sha-02c62e46ac16` tag rather than `latest`. Measured reason: the
#91 and #93 builds started 25 seconds apart and ran concurrently, both pushing `latest`, and
for iac-toolbox the earlier run's push landed last. `latest` resolved to 7f5292725111 while
the 774ab22 build had published c1d133bb0e97. The per-commit tag is race-free. build-toolbox.yml
has no concurrency guard, which is a separate thing to fix.

Each digest was verified before being written: its build-inputs label was read back out of the
registry and compared with the fingerprint this repo computes now, and nothing would have been
written if any of the five disagreed. scripts/check-image-pins.py then reports all five current.
garymike added a commit that referenced this pull request Oct 30, 2026
The gate went red again four days after #91 cleared it; four releases landed upstream in
that window.

| Tool | From | To |
|---|---|---|
| snyk-agent-scan | 0.6.9 | 0.6.10 |
| semgrep | 1.182.1 | 1.183.0 |
| checkov | 3.3.30 | 3.4.0 |
| skillspector | 2.13.0 | 2.13.2 |

The first three went through scripts/apply-tool-bumps.py, so tools.lock and the matching
Dockerfile ARGs moved together. skillspector is a hand bump by design: its ARG is a commit
SHA rather than a version, and that script refuses it on purpose, because patch_lock would
rewrite the lock while patch_dockerfiles matched nothing and the half-bump would report as
applied. v2.13.2 = abd82935b67f. check-tool-updates.sh --fail-on-drift now exits 0 on all
twelve pins.

Re-measured every fixture at 2.13.2, which is what the drift warning for that pin asks for.
Nothing moved: gecko-demo 64/100, gecko-hook-demo 76/100, config-injection-demo 100/100,
sibling-config-demo 51/100, and the persistent-memory write still cleared at 0/100.
gate-proof.sh passes 16 of 16 against a freshly built image carrying 2.13.2 and semgrep
1.183.0.

v2.13.2 predates NVIDIA/SkillSpector#893, checked against the commit graph rather than
inferred from release dates: the merge commit is 1196 commits ahead of that tag. So the
agent-memory write is still the carrier this gate uniquely covers, and the 0/100 figure in
README.md, docs/ and the Dockerfiles stays accurate. The release that first carries #893 is
the one that moves that fixture to 27/100 and makes those references stale.

Also corrects the PyJWT override's comment in both Dockerfiles. It claimed semgrep pins
pyjwt[crypto]~=2.13.0; as of 1.183.0 semgrep requires <3,>=2.15.0, so this bump made the
comment false. The override itself is left alone: removing it un-pins pyjwt in a repo that
pins everything deliberately, and that trade deserves its own change.
garymike added a commit to garymike/security-agents that referenced this pull request Oct 30, 2026
Was 79b45d2, seven commits behind. This picks up the tool bumps in garymike/security-workflows#91
and #96 (trufflehog 3.99.7, snyk-agent-scan 0.6.10, semgrep 1.183.0, checkov 3.4.0,
skillspector 2.13.2), the corrected agent-memory destination rule from #93, and the rebuilt
image-pin check from #94. #97 moved the five toolbox image digests, so a caller at bee631e
resolves the current toolchain rather than the pre-#91 one.

The rule fix matters here specifically: before #93 a write to MEMORY.md through a raw-string or
f-string path produced no finding, which is the shape NVIDIA/SkillSpector#893 was sent back in
review for missing. A caller pinned behind that fix does not catch it.

Dates in the trailing comments corrected too. They read 2026-10-21 against a SHA that moved on
2026-10-30, because a bump that changes the ref and leaves the comment makes the comment a lie.
garymike added a commit to garymike/skills that referenced this pull request Oct 30, 2026
Was 79b45d2, seven commits behind. This picks up the tool bumps in garymike/security-workflows#91
and #96 (trufflehog 3.99.7, snyk-agent-scan 0.6.10, semgrep 1.183.0, checkov 3.4.0,
skillspector 2.13.2), the corrected agent-memory destination rule from #93, and the rebuilt
image-pin check from #94. #97 moved the five toolbox image digests, so a caller at bee631e
resolves the current toolchain rather than the pre-#91 one.

The rule fix matters most for this repo, since the skill-audit gate is what guards the skills
here: before #93 a write to MEMORY.md through a raw-string or f-string path produced no finding,
which is the shape NVIDIA/SkillSpector#893 was sent back in review for missing. SkillSpector
2.13.2 still clears that carrier at 0/100 (it predates #893), so the first-party gate is the
only thing catching it, and until now this repo resolved the version that could not.

Dates in the trailing comments corrected too. They read 2026-10-19 against a SHA from
2026-10-21, because #36 moved the ref and left the comment alone.
garymike added a commit that referenced this pull request Oct 30, 2026
… at once (#98)

build-toolbox.yml had no concurrency guard, and that was not a theoretical gap: the #91 and
#93 builds ran 25 seconds apart, both pushed `latest`, and for iac-toolbox the EARLIER run's
push landed last. `latest` named the older build while the newer digest existed only under
its `sha-` tag.

The group omits github.ref deliberately. The registry tags are global, so a
workflow_dispatch from another branch has to queue behind main rather than get its own lane.

cancel-in-progress is false, matching tool-bump.yml and for the same reason. These jobs push
and sign six images; killing a run mid-flight can leave the base published and the five
domain images not, which is a worse state than a slower queue. Nothing in flight is
interrupted, runs serialize, and the newest commit's run finishes last so it wins `latest`.
GitHub keeps one pending run per group, so a burst of pushes collapses to the one running
plus the newest, which is the right pair to keep.

Found while moving the image pins, by comparing what the registry served against what the
build log said it published, rather than trusting either on its own. Nothing broke because of
it: scripts/check-image-pins.py and the pin mover both resolve the per-commit `sha-` tag
precisely so a raced `latest` cannot be pinned. `latest` is still what a human pulling the
image by tag gets, which is why this is worth fixing rather than routing around.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant