Commit 4d68b36
authored
ci: rename Claude OAuth secret to CC_OAUTH_TOKEN (#702)
## Summary
- Switch `.github/workflows/claude.yml` to read the OAuth credential
from `secrets.CC_OAUTH_TOKEN` instead of
`secrets.CLAUDE_CODE_OAUTH_TOKEN`. Same token; new secret name.
- The action input name (`claude_code_oauth_token`) is unchanged.
## Why
Same workflow file, same `claude-code-action` SHA (`11a9dadd` =
v1.0.108), same agent SDK (`0.2.121`), and an unchanged secret value
produced **6 consecutive successful `@claude` runs (15:27–15:55 UTC)**
and then **4 consecutive failures (16:00–16:13 UTC)** on 2026-04-28, all
with the SDK error `Could not resolve [authentication]` thrown from
`validateHeaders`. Failing run for reference:
https://github.com/bootjp/elastickv/actions/runs/25063998905.
Successful and failing runs both reach `App token successfully obtained`
in the OIDC → GitHub App exchange, and both download the same action SHA
+ same SDK version. The crash is in the agent SDK's `Anthropic` client
constructor: neither `apiKey` nor `authToken` resolves. Workflow-side
variables are exhausted, so the remaining moving parts are the secret
name itself and the OAuth token behind it.
The previous name `CLAUDE_CODE_OAUTH_TOKEN` collides with the env var
that Claude Code itself recognizes as a credential, so any process in
the pipeline that touches its own login state (the action's child
`claude` process, the runner-side `claude` install, or developer-machine
tooling that shares the same env can clobber, mask, or otherwise
interact with that env. Renaming the GitHub Actions secret avoids the
magic-name interaction.
## Behavior change
- Active only when is mentioned. Other workflows are untouched.
- Token has been re-registered under with a freshly issued OAuth token;
the old secret remains in place but is no longer referenced.
## Risk
- If the new secret is missing or empty, invocations will fail at the
same point as today's failures (no regression versus the failing
baseline).
- No code path / data-plane / replication / consistency surface — CI
workflow only.
## Test plan
- [ ] After merge, comment on this PR (or any open PR) and observe a
successful run.
- [ ] If still failing with the same error, the next step is to re-add
the explicit forwarding (the workaround) and/or pin to v1.0.107 () until
upstream fixes the auth-forwarding regression. That follow-up is
intentionally out of scope here so this PR cleanly tests the magic-name
hypothesis in isolation.
## Self-review
5-lens collapses to no-op: CI workflow change only, no data-loss /
concurrency / performance / consistency surface, no test-coverage
applicable.
PR_EOF
)1 file changed
Lines changed: 1 addition & 1 deletion
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
34 | 34 | | |
35 | 35 | | |
36 | 36 | | |
37 | | - | |
| 37 | + | |
38 | 38 | | |
39 | 39 | | |
40 | 40 | | |
| |||
0 commit comments