fix(ci): remove a keyword GitHub rejects on reusable-workflow callers

`continue-on-error` is not allowed on a job that calls a reusable workflow.
GitHub answers with "Property continue-on-error is not allowed" and refuses the
**whole file** — a startup_failure with zero jobs and no log, not one failed
job. Confirmed against two GitHub community discussions, one of which is an
open, unanswered feature request asking for the capability.

release.yml's announce job has carried that keyword since 5f4c6e7. That commit
landed *after* the v1.0.0 tag object and no tag has been pushed since, so the
first v1.0.1 tag would have been its first execution — and the release would
have died before running a single job, in exactly the silent shape this project
already hit once with the `contents: write` grant (2c8fba9). The previous commit
in this branch then copied the same mistake into notify-ci-failure.yml.

Both are now gated with `if:` on a repo variable instead. A variable and not a
secret because secrets cannot be referenced in a job-level `if:`, and unset
means skipped — so the release stays green and CI stays quiet until the owner
opts in by setting DISCORD_ANNOUNCE / DISCORD_CI_NOTIFY alongside the webhooks.
Same idiom as the CWS_AUTOPUBLISH gate already in release.yml.

docs/12's samples and its §5 checklist updated in the same commit; docs are
binding here, and leaving the invalid keyword in the spec would have had the
next reader put it straight back.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
poli0981andClaude Opus 5 committed 2026-08-03 20:20:05 +07:00
1 parent f3d4dfb023
commit dc0672ea28
4 files changed
+66 -26

No files matched your search

+15 -6
View File
@@ -7,10 +7,10 @@
# unset (docs/00 §10.6) this workflow therefore failed on *every* CI completion,
# success included, from the first push after it landed.
#
# `continue-on-error` on top of that, for the same reason `release.yml`'s
# announce job has it: a notification must never be able to change the colour of
# the thing it is reporting on. A run that is red for a reason nobody acts on
# teaches everyone to stop reading the colour.
# The `vars.DISCORD_CI_NOTIFY` gate on top of that, for the same reason
# `release.yml`'s announce job has one: a notification must never be able to
# change the colour of the thing it is reporting on. A run that is red for a
# reason nobody acts on teaches everyone to stop reading the colour.
name: Notify CI failure
on:
@@ -24,8 +24,17 @@ permissions:
jobs:
notify:
if: github.event.workflow_run.conclusion == 'failure'
continue-on-error: true
# Both conditions matter. The conclusion check stops this running at all on
# a green CI, which is what made it red on every push; the variable gate
# keeps it off entirely until the owner has actually set DISCORD_CI_WEBHOOK,
# since the reusable declares that secret `required: true` and GitHub
# validates it before the reusable's own filter runs.
#
# `continue-on-error` is NOT usable here: GitHub rejects it on a job that
# calls a reusable workflow ("Property continue-on-error is not allowed"),
# which invalidates the entire file. A repo variable is allowed in `if:`;
# a secret is not.
if: github.event.workflow_run.conclusion == 'failure' && vars.DISCORD_CI_NOTIFY == 'true'
uses: poli0981/.github/.github/workflows/notify-ci-failure.yml@main
permissions:
contents: read
+14 -5
View File
@@ -90,13 +90,22 @@ jobs:
# would silently never fire.
announce:
needs: release
# A Discord notification is not part of the release. The reusable exits 1
# on "No webhooks configured" when DISCORD_RELEASES_WEBHOOK and
# A Discord notification is not part of the release. The reusable exits 1 on
# "No webhooks configured" when DISCORD_RELEASES_WEBHOOK and
# DISCORD_REPO_WEBHOOK are both unset (docs/12 §5 lists them as optional),
# which turned the v1.0.0 run red even though the Release, the zip and
# SHA256SUMS.txt had all published correctly. A red run on a good release
# is worse than a missing announcement: it trains you to ignore the colour.
continue-on-error: true
# SHA256SUMS.txt had all published correctly. A red run on a good release is
# worse than a missing announcement: it trains you to ignore the colour.
#
# ⚠️ This used to say `continue-on-error: true`, which GitHub REJECTS on a
# job that calls a reusable workflow — "Property continue-on-error is not
# allowed". The whole file is then invalid and the run is a startup_failure
# with zero jobs and no log, exactly like the `contents: write` mistake in
# 2c8fba9. It was never caught because that fix landed after the v1.0.0 tag
# and no tag has been pushed since, so the first v1.0.1 tag would have been
# its first execution. Gate on a repo VARIABLE instead — allowed in `if:`,
# unlike a secret, and the same idiom as CWS_AUTOPUBLISH above.
if: vars.DISCORD_ANNOUNCE == 'true'
uses: poli0981/.github/.github/workflows/announce-release.yml@main
permissions:
contents: read
+36 -15
View File
@@ -220,7 +220,7 @@ jobs:
announce:
needs: release
continue-on-error: true # see below
if: vars.DISCORD_ANNOUNCE == 'true' # see below — NOT continue-on-error
uses: poli0981/.github/.github/workflows/announce-release.yml@main
permissions:
contents: read
@@ -236,11 +236,25 @@ tag: a re-run could not get past that line, and recovery meant deleting the
Release and the tag by hand. `docs/13 §7` documents the manual escape; this
removes the need for it in the common case.
`continue-on-error` because the reusable exits 1 on "No webhooks configured"
when both Discord secrets are unset — which §5 lists as optional. On the real
v1.0.0 run the Release, the zip and `SHA256SUMS.txt` all published correctly
and the run still showed red. A red run on a good release trains you to ignore
the colour, so the notification must not be able to fail the release.
The announcement is gated because the reusable exits 1 on "No webhooks
configured" when both Discord secrets are unset — which §5 lists as optional. On
the real v1.0.0 run the Release, the zip and `SHA256SUMS.txt` all published
correctly and the run still showed red. A red run on a good release trains you
to ignore the colour, so the notification must not be able to fail the release.
⚠️ **It is gated with `if:` on a repo VARIABLE, not with `continue-on-error`.**
GitHub **rejects** `continue-on-error` on a job that calls a reusable workflow —
`Property continue-on-error is not allowed` — and a rejected file does not fail
one job, it makes the whole run a `startup_failure` with zero jobs and no log.
That is the same silent shape as the `contents: write` mistake in §2.1, and it
sat undetected in this repo from `5f4c6e7` until 2026-08-03 precisely because
that commit landed *after* the `v1.0.0` tag and no tag was pushed in between —
the first v1.0.1 tag would have been its first execution.
A **variable** and not a secret: secrets cannot be referenced in a job-level
`if:`. `vars.DISCORD_ANNOUNCE` defaults to unset, so the announcement is skipped
and the release stays green until the owner opts in — the same idiom as
`CWS_AUTOPUBLISH`. Set the variable only once the webhook secrets exist.
`tag_override` is **required here**, and for the same reason the announcement
is a job rather than a caller. The reusable resolves
@@ -269,8 +283,7 @@ permissions:
jobs:
notify:
if: github.event.workflow_run.conclusion == 'failure'
continue-on-error: true
if: github.event.workflow_run.conclusion == 'failure' && vars.DISCORD_CI_NOTIFY == 'true'
uses: poli0981/.github/.github/workflows/notify-ci-failure.yml@main
permissions:
contents: read
@@ -285,14 +298,17 @@ cost, and not for the failure below.
⚠️ **The reusable declares `DISCORD_CI_WEBHOOK` as `required: true`**, which
GitHub validates before any of the reusable's own logic runs. With the secret
unset (`00 §10.6`), the caller therefore failed on **every** CI completion —
success or failure — from the first push after this file landed. The `if:` stops
it being called on green runs at all, and `continue-on-error` keeps a missing
optional webhook from turning a genuine CI *failure* into two failures.
success or failure — from the first push after this file landed.
This is the same reasoning as `2.3`'s `continue-on-error` on `announce`, learned
the same way: a notification must never be able to change the colour of the
thing it is reporting on, because a run that is red for a reason nobody acts on
teaches everyone to stop reading the colour.
Both halves of the condition earn their place: the conclusion check stops it
running on green CI at all, and `vars.DISCORD_CI_NOTIFY` keeps it off entirely
until the secret actually exists. **Not `continue-on-error`** — see §2.3: GitHub
rejects that keyword on a reusable-workflow caller and invalidates the file.
Same reasoning as §2.3's announce gate, learned the same way: a notification
must never be able to change the colour of the thing it is reporting on, because
a run that is red for a reason nobody acts on teaches everyone to stop reading
the colour.
## 3. Chrome Web Store publishing automation
@@ -345,6 +361,11 @@ means "unknown", not "not done".
- [ ] ⚠️ Repo variable `CWS_AUTOPUBLISH=false` initially. Note that when the
variable does not exist the expression is simply false, so "not configured"
and "configured off" are indistinguishable from a run's output.
- [ ] ⚠️ Repo variables `DISCORD_ANNOUNCE` and `DISCORD_CI_NOTIFY`. Leave both
unset until the matching webhook secrets exist — the reusables declare
those secrets required, so calling them without one fails the job, and a
reusable-workflow caller cannot use `continue-on-error` to absorb it
(§2.3). Unset means skipped, which is the safe default.
- [x] Dependabot: npm weekly, grouped dev minor+patch, `github-actions` too —
`.github/dependabot.yml`. Note it has no awareness of the `package.json`
`overrides` block, so those six pins must be reviewed by hand (`09 §3.1`).
+1
View File
@@ -203,5 +203,6 @@ independently of the package (`13 §5`), though the copy should still go through
| 2026-08-03 | **The "if you contact us" paragraph in `legal/PRIVACY.md` is a clarification, not a material change** — it describes what happens to data a user chooses to send, which was never inside "what the extension collects". `LEGAL_VERSION` deliberately **not** bumped: re-showing the gate to every installed user over a patch is a worse outcome than the ambiguity it fixes | 14 §2, 14 §3 |
| 2026-08-03 | **A WXT bump is accepted only after diffing the BUILT manifest**, not on a green test run. `tests/unit/manifest.test.ts` reads `wxt.config.ts`, which a WXT upgrade cannot change, so it stays green no matter what WXT emits. 0.21.3 was verified byte-identical in manifest, file inventory and bundle size; it did break `wxt/testing` into two subpaths and retype `tabs.create` as `void`, both dev-only. `check-manifest.mjs` now asserts the exact top-level key set so the next one is caught by CI rather than by hand | 01 §2, 12 §2.3 |
| 2026-08-03 | **jsdom 30 accepted after running `dom-environment.test.ts` first.** The canary exists because the Netscape walk depends on parse5's implied-end-tag behaviour; the rule is that a failing canary pins the dependency, never loosens the test. The 2 869 ms figure in `11 §6` is a jsdom **29** number and a major bump adds a column rather than overwriting it | 01 §2, 11 §6 |
| 2026-08-03 | **`continue-on-error` is invalid on a job that calls a reusable workflow** — GitHub rejects the file with "Property continue-on-error is not allowed", and a rejected file is a `startup_failure` with zero jobs and no log, not a failed job. `release.yml`'s announce job had carried it since `5f4c6e7`; that commit landed *after* the `v1.0.0` tag and no tag was pushed since, so **the first v1.0.1 tag would have been its first execution**. Both notification jobs are now gated with `if:` on a repo variable — allowed there, unlike a secret — defaulting to skipped | 12 §2.3, 12 §2.4 |
| 2026-08-03 | **A notification must never change the colour of the thing it reports on.** `notify-ci-failure.yml` had been failing on *every* CI completion, success included, because the reusable declares a secret that does not exist and GitHub validates that before the reusable's own success filter runs. Same lesson as `5f4c6e7`, learned twice | 12 §2.4 |
| 2026-08-03 | **No support email on any in-product surface.** About links to the store listing, the repo, issues and the changelog. Publishing an address inside the extension invites unsanitized bookmark files by mail, which is the exact receipt-of-personal-data problem the privacy clause exists to bound. The address stays in `SECURITY.md`, `CONTRIBUTING.md` and the privacy policy | 06 §3.5, 09 §6 |