Repository navigation
feat!: ci may not skip a push verdict - #19
Merged
Merged
Conversation
There was a problem hiding this comment.
Gatehouse found no issues.
Reviewed 1 commit since b952ec3 (incremental).
This was referenced Nov 5, 2026
There was a problem hiding this comment.
Gatehouse found no issues.
Reviewed 1 commit since 248c478 (incremental).
This was referenced Nov 5, 2026
Closed
Merged
Constitution XII says CI runs push's checks except Gourmand and Gatehouse. Those two are CI_EXCLUDES, decided here. Skipping any other verdict in the `ci` block alone left it on pre-push -- a developer's machine, one `--no-verify` from gone -- while the required `sdlc ci` reported green. 24 of the 33 repos converted under RT#1552 carried `[tool.sdlc.ci] skip = ["codeql"]`, each justified by a comment saying the CodeQL Action covered it. Against the rulesets: that check was required in 1, advisory in 21, and never ran on a pull request in 2 -- ashigaru has no CodeQL workflow at all, and rotv's upload-sarif step is gated on `github.event_name != 'pull_request'`. So in 23 repos the only place CodeQL could block was a hook I had been pushing past with --no-verify. resolve() now refuses the config and names what it objects to. sdlc_config() validates every block whichever command runs, so a repo that adds the skip back finds out on its next `sdlc commit` rather than in CI -- where the skip is exactly what would have made it read green. A check the repo added itself may still be skipped in ci; a check that should run nowhere belongs out of [tool.sdlc.push], where the decision sits in one place. The TRENTINA fixture in tests/test_plan.py carried the skip and now does not; six new refusal cases cover every verdict, plus the two that stay legal, a repo-added check, and the commit-time catch. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R7Y1jesjfgxHeqtKJCvay6
Crunchtools 1's call, and the right one: 0.4.0 is ready now and this is blocked behind 24 consumer repos dropping the ci codeql skip, so holding the number here would park the fleet behind that. Rebases onto main once 0.4.0 merges. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01R7Y1jesjfgxHeqtKJCvay6
fatherlinux
force-pushed
the
feature/ci-may-not-skip-a-verdict
branch
from
November 5, 2026 02:06
2b24ade to
5cf69ce
Compare
fatherlinux
enabled auto-merge (squash)
November 5, 2026 02:07
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
[tool.sdlc.ci] skipmay no longer name a check the push tier renders a verdict with — mypy, pytest, semgrep, pip-audit, constitution, codeql.resolve()refuses the config and names what it objects to.Why
Constitution XII: "CI runs push's checks except Gourmand and Gatehouse." Those two are
CI_EXCLUDES, decided by this tool, not by a repo. Skipping any other verdict in theciblock alone leaves it on pre-push — a developer's machine, one--no-verifyfrom gone — while the requiredsdlc cireports green.That is not hypothetical. 24 of the 33 repos converted under RT#1552 carried
[tool.sdlc.ci] skip = ["codeql"], each with a comment claiming the CodeQL Action covered it. Checked against the rulesets:CodeQL Analysisis a required checkashigaruhas no CodeQL workflow;rotv'supload-sarifis gated ongithub.event_name != 'pull_request'So in 23 of them the only place CodeQL could block was a hook I had been pushing past with
--no-verify. I wrote those skips; this is the fix.All 24 consumer repos have now dropped the skip and merged — 0 of 33
ciblocks skip a verdict. Turning CodeQL on found real findings in 4 of them straight away, including two credential exposures in rotv'sscripts/get_google_token.py(an OAuth refresh token written world-readable, and all three credentials printed to stdout), both fixed.Why this one goes second
0.4.0 (#22) had to land first, and not only for version ordering. #22 dropped all capabilities from the gate's own containers, and that broke CodeQL: the bundle's 25 precompiled
.qlxfiles extract at mode 0600 owned by uid 1001, and root had been reading them purely onCAP_DAC_OVERRIDE. Without it, every query recompiled andcodeql database analyzedied with exit 100 — 355 seconds and an exit 2 instead of a verdict. #22 fixed it in the image (chown/chmodafter extraction rather than handing the capability back, which also took codeql from 355s to 95s).Merged the other way round, the 24 repos would have dropped their
ciskip just in time for CodeQL to stop running everywhere. Worth stating plainly, because this PR's whole claim is that CodeQL can now block, and that claim depends on #22.Where the refusal lands
sdlc_config()validates every block whichever command runs, so a repo that adds the skip back finds out on its nextsdlc commit, not in CI — which matters, because in CI the skip is exactly what would have made it read green.Still legal: skipping a check the repo added itself, and naming
gourmand/gatehouse(which ci never runs anyway). A check that should run nowhere belongs out of[tool.sdlc.push], where the decision is visible in one place rather than as a difference between two tiers nobody reads side by side.Verification
tests/test_plan.py: one case per verdict asserting the refusal, generated fromDEFAULTS["push"] - CI_EXCLUDESso it grows by itself if a verdict is added; plus the two exclusions staying legal, a repo-added check still skippable, and the commit-time catch. TheTRENTINAfixture carried the skip and no longer does. Full run against the rebased branch: all passed.🤖 Generated with Claude Code
https://claude.ai/code/session_01R7Y1jesjfgxHeqtKJCvay6