Conversation
The credential screen on the `GITHUB_`/`EVALS_` prefix rules reads one
segment:
!CREDENTIAL_SUFFIXES.has(k.slice(k.lastIndexOf('_') + 1).toUpperCase())
A trailing qualifier moves the credential word off the end, and the name is
admitted again. That is the ordinary spelling once there is more than one of
something, so these all reach the hermetic child today:
GITHUB_APP_PRIVATE_KEY_BASE64 the PEM, base64-encoded
GITHUB_PRIVATE_KEY_PEM the PEM
GITHUB_TOKEN_1 a numbered token
GITHUB_TOKEN_GHES an enterprise-server token
GITHUB_CLIENT_SECRET_VALUE an OAuth app secret
`isCredentialShapedName` now tests every underscore-separated segment.
Segments rather than substrings, because `GITHUB_PATH` is documented runner
metadata and contains PAT, and `GITHUB_TOKENIZER` and `GITHUB_KEYRING` are
already pinned as names that must survive.
The screen still governs prefix rules only: an exact `ALLOW_EXACT` name and a
runner's `extraAllow` re-admit their credentials as before.
Checked against the full documented `GITHUB_*` runner set plus the `EVALS_`
knobs: none of them has a credential word in any segment, so nothing that
was metadata stops being metadata.
|
Merging to
After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here |
garrytan
added a commit
that referenced
this pull request
Oct 3, 2026
…g/upgrade/type --selector, BROWSER.md type and profiles, AGENTS.md conventions, CONTRIBUTING #2949, touchfiles for the wave's new lib/bin files
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up to the credential screen adapted from #2636 and merged as
b9706f3(#2942, v1.88.1.0) — thank you for the credit there. One residual hole in it.The hole
The screen reads a single segment:
A trailing qualifier moves the credential word off the end and the name is admitted again. That qualifier is the ordinary spelling once there is more than one of something. Running the merged
buildHermeticEnvonb9706f3:GITHUB_APP_PRIVATE_KEY_BASE64is the same PEM the screen already rejects under its unqualified name, just base64-encoded — which is how a GitHub App key is normally carried in an environment variable in the first place.The change
isCredentialShapedNametests every underscore-separated segment.Segments rather than substrings, deliberately:
GITHUB_PATHis documented runner metadata and containsPAT, andGITHUB_TOKENIZERandGITHUB_KEYRINGare already pinned in the suite as names that must survive. A substring screen takes all three; the mutation table below pins that.The screen still governs prefix rules only. An exact
ALLOW_EXACTname and a runner'sextraAllowre-admit their credentials exactly as before —GEMINI_*carryingGEMINI_API_KEYis unchanged, and there is a new test for the qualified form of that opt-in.Checked against the full documented
GITHUB_*runner set (the 33 names your own test enumerates) plus theEVALS_knobs: no segment of any of them is a credential word, so nothing that was metadata stops being metadata.CREDENTIAL_SUFFIXESis renamedCREDENTIAL_SEGMENTS, since "suffix" is no longer what it means.Coverage
Four tests added to
test/helpers/hermetic-env.test.ts, and the launched-child regression intest/helpers/session-runner.test.tsgrows two credential names and one near-miss metadata name — a predicate test alone would not have shown this reaching a child.extraAllowtoobun test test/helpers/hermetic-env.test.ts: 33 pass. Three failures in that file are Windows-only and identical with the change stashed (split CEO artifact Read scope×2,Design artifact Read scope— EPERM onsymlink).One thing I could not run here. The launched-child regression cannot execute on Windows on either side of this change: the fixture
claudeis written with a#!/usr/bin/env nodeshebang, so Windows cannot launch it and the child's stdout is a shell error rather than JSON. I verified the table I added to it by drivinghermeticChildEnv()— the same builder that test's child env comes from — over the same names, and all eleven match the asserted values. CI runs the real thing.