fix(mirror): decouple Codeberg mirror via release:published event (#427) #428

Merged

quartermaster merged 2 commits from i/427-mirror-decouple-release-published into main 2026-07-06 10:38:32 +02:00

Conversation 2 Commits 2 Files changed 4 +111 -48

quartermaster commented 2026-07-06 10:24:22 +02:00

Owner

Copy link

Summary

v1.0.0 quality per operator ratify (Bosun dispatch 05f3, my architectural lean in #427 body). Decouples the Codeberg mirror from the cut workflow's job graph via the release: published event, resolving Forgejo Actions' schema-validator errors on .forgejo/workflows/release.yml:80-97 by construction. Closes #427.

Root cause + resolution

continue-on-error is not valid on uses: (workflow_call) jobs per the GitHub Actions schema — only runs-on: jobs support it. GitHub Actions accepts + ignores it silently; Forgejo Actions' validator correctly flags it. Non-fatal at runtime, but polluted Codeberg UI + IDE tooling.

Decoupled architecture removes the coupling entirely: no continue-on-error gymnastics because mirror-failure never touches the cut workflow's job graph.

Substrate delta

• New workflow file .forgejo/workflows/mirror-on-cut.yml:
  • on: release: types: [published] — fires when a release transitions to published (either immediate-mode straight from draft-release.sh, or draft-mode when the operator clicks Publish in the Forgejo UI)
  • Calls the same reusable-mirror-to-codeberg.yml as before
  • tag_name: ${{ github.event.release.tag_name }} threads the release-event payload
  • secrets: inherit for CODEBERG_MIRROR_PAT
• release.yml: removed the in-graph mirror: job (with needs, if, continue-on-error, uses, with, secrets). Kept a short comment block pointing at mirror-on-cut.yml for the rationale.
• reusable-mirror-to-codeberg.yml consumer-wiring docstring: updated from the coupled jobs: release: ..., mirror: needs: release ... pattern to the new event-decoupled on: release: types: [published] pattern.
• reusable-mirror-to-codeberg.yml Design section (post Surveyor 3648 catch): reconciled continue-on-error: true MUST-mandate that had been left stale after the Consumer-wiring block was updated. Same grep-sweep-missed-siblings shape as #406's required→recommended sweep — when reframing a concept, sweep the FULL surface. grep -n continue-on-error now returns zero prescriptive refs (the surviving four are all descriptive: two in the top-of-file decoupling context, two in the new #427 explainer).

Substrate benefits

• Design-intent alignment: "mirror is downstream visibility, not source-of-record" — the coupling belongs at the release-publish boundary, not inside the cut workflow's job graph. Loose coupling encodes this cleanly.
• Schema errors resolve by construction: no more uses: + continue-on-error combo in release.yml, so Forgejo's validator has nothing to complain about.
• Mirror failure never cascades into cut-failure by construction: same principle continue-on-error was trying to encode, achieved via event-decoupling instead of in-graph error suppression. Any future mirror-side change (timeout policy, retry logic) can iterate without touching the cut workflow.
• Latent draft-mirror bug fixed incidentally (Surveyor 3648 flag): the prior in-graph mirror: job fired on mode==cut regardless of publish_mode. In draft-mode that meant we mirrored an UNPUBLISHED draft (created but not yet published). The release:published trigger correctly waits for actual publish — the decoupling fixes this latent bug too.

Verification

• tests/workflows.bats: 18/18 green. No existing guard covered the mirror-job specifically, so removing it doesn't fail any regression test. The schema-validator issue itself doesn't have a bats fixture — it's a live Forgejo Actions surface.
• Full sweep: 679/679 EXIT=0 (count-verified per feedback_bats_sweep_count_verification).
• Register-check clean at HEAD (verified post-staging per feedback_verify_after_mutation).
• grep -n continue-on-error .forgejo/workflows/reusable-mirror-to-codeberg.yml: 4 remaining refs, all descriptive — zero prescriptive MUST/SHALL. Surveyor's grep-sweep pattern applied.
• Forgejo event-decoupling — high-confidence-plus-verify (Surveyor 3648 de-risking): tmux-msg/.forgejo/workflows/release-publish.yml uses the IDENTICAL on: release: types: [published] shape on the same Forgejo 15.0.2 / gitea-1.22.0 instance. In-ecosystem precedent exists — not a leap of faith. Definitive test remains v0.28.0's first live cut (post-operator's v0.27.0 recovery); if release:published doesn't fire, Bug #2 recurs and I iterate on the tracker.

Verification AC (from tracker)

• .forgejo/workflows/release.yml passes Forgejo schema validation cleanly (the offending block is gone)
• Mirror-on-cut behavior preserved via the new workflow — same reusable-mirror-to-codeberg.yml, same params, just event-triggered
• Mirror failure doesn't block cut path (existing invariant, now enforced by architecture instead of continue-on-error)
• No regression on tests/workflows.bats guards
• Reusable docstring Design section reconciled with event-decoupled architecture (post-Surveyor 3648)

Files

• New: .forgejo/workflows/mirror-on-cut.yml (event-triggered wrapper), changelog.d/427.fixed.md
• Modified: .forgejo/workflows/release.yml (mirror job removed, comment kept), .forgejo/workflows/reusable-mirror-to-codeberg.yml (Consumer-wiring + Design section updated to event-decoupled pattern)

Refs #427 (main tracker), Bosun dispatch fff2 (Bug #2 of three, PR#406 empirical anchor), Bosun dispatch 05f3 (operator ratify of Option B), Surveyor review 3648 @ 5570a7f (grep-sweep-missed-siblings docstring catch). Sibling v1.0.0-quality gates: #417 (orphan-check false-positive) + #420 (paragraph normalization) — same arc.

## Summary **v1.0.0 quality** per operator ratify (Bosun dispatch 05f3, my architectural lean in #427 body). Decouples the Codeberg mirror from the cut workflow's job graph via the `release: published` event, resolving Forgejo Actions' schema-validator errors on `.forgejo/workflows/release.yml:80-97` **by construction**. Closes #427. ## Root cause + resolution `continue-on-error` is not valid on `uses:` (workflow_call) jobs per the GitHub Actions schema — only `runs-on:` jobs support it. GitHub Actions accepts + ignores it silently; Forgejo Actions' validator correctly flags it. Non-fatal at runtime, but polluted Codeberg UI + IDE tooling. **Decoupled architecture** removes the coupling entirely: no `continue-on-error` gymnastics because mirror-failure never touches the cut workflow's job graph. ## Substrate delta - **New workflow file** `.forgejo/workflows/mirror-on-cut.yml`: - `on: release: types: [published]` — fires when a release transitions to published (either `immediate`-mode straight from `draft-release.sh`, or `draft`-mode when the operator clicks Publish in the Forgejo UI) - Calls the same `reusable-mirror-to-codeberg.yml` as before - `tag_name: ${{ github.event.release.tag_name }}` threads the release-event payload - `secrets: inherit` for `CODEBERG_MIRROR_PAT` - **`release.yml`**: removed the in-graph `mirror:` job (with `needs`, `if`, `continue-on-error`, `uses`, `with`, `secrets`). Kept a short comment block pointing at `mirror-on-cut.yml` for the rationale. - **`reusable-mirror-to-codeberg.yml`** consumer-wiring docstring: updated from the coupled `jobs: release: ..., mirror: needs: release ...` pattern to the new event-decoupled `on: release: types: [published]` pattern. - **`reusable-mirror-to-codeberg.yml`** Design section (post Surveyor 3648 catch): reconciled `continue-on-error: true` MUST-mandate that had been left stale after the Consumer-wiring block was updated. Same **grep-sweep-missed-siblings** shape as #406's `required→recommended` sweep — when reframing a concept, sweep the FULL surface. `grep -n continue-on-error` now returns zero prescriptive refs (the surviving four are all descriptive: two in the top-of-file decoupling context, two in the new #427 explainer). ## Substrate benefits - **Design-intent alignment**: "mirror is downstream visibility, not source-of-record" — the coupling belongs at the release-publish boundary, not inside the cut workflow's job graph. Loose coupling encodes this cleanly. - **Schema errors resolve by construction**: no more `uses:` + `continue-on-error` combo in `release.yml`, so Forgejo's validator has nothing to complain about. - **Mirror failure never cascades into cut-failure by construction**: same principle `continue-on-error` was trying to encode, achieved via event-decoupling instead of in-graph error suppression. Any future mirror-side change (timeout policy, retry logic) can iterate without touching the cut workflow. - **Latent draft-mirror bug fixed incidentally** (Surveyor 3648 flag): the prior in-graph `mirror:` job fired on `mode==cut` regardless of `publish_mode`. In `draft`-mode that meant we mirrored an UNPUBLISHED draft (created but not yet published). The `release:published` trigger correctly waits for actual publish — the decoupling fixes this latent bug too. ## Verification - **`tests/workflows.bats`**: 18/18 green. No existing guard covered the mirror-job specifically, so removing it doesn't fail any regression test. The schema-validator issue itself doesn't have a bats fixture — it's a live Forgejo Actions surface. - **Full sweep**: **679/679 EXIT=0** (count-verified per `feedback_bats_sweep_count_verification`). - **Register-check clean at HEAD** (verified post-staging per `feedback_verify_after_mutation`). - **`grep -n continue-on-error .forgejo/workflows/reusable-mirror-to-codeberg.yml`**: 4 remaining refs, all descriptive — zero prescriptive MUST/SHALL. Surveyor's grep-sweep pattern applied. - **Forgejo event-decoupling — high-confidence-plus-verify** (Surveyor 3648 de-risking): tmux-msg/.forgejo/workflows/release-publish.yml uses the IDENTICAL `on: release: types: [published]` shape on the same Forgejo 15.0.2 / gitea-1.22.0 instance. In-ecosystem precedent exists — not a leap of faith. Definitive test remains v0.28.0's first live cut (post-operator's v0.27.0 recovery); if `release:published` doesn't fire, Bug #2 recurs and I iterate on the tracker. ## Verification AC (from tracker) - [x] `.forgejo/workflows/release.yml` passes Forgejo schema validation cleanly (the offending block is gone) - [x] Mirror-on-cut behavior preserved via the new workflow — same `reusable-mirror-to-codeberg.yml`, same params, just event-triggered - [x] Mirror failure doesn't block cut path (existing invariant, now enforced by architecture instead of `continue-on-error`) - [x] No regression on `tests/workflows.bats` guards - [x] Reusable docstring Design section reconciled with event-decoupled architecture (post-Surveyor 3648) ## Files - **New**: `.forgejo/workflows/mirror-on-cut.yml` (event-triggered wrapper), `changelog.d/427.fixed.md` - **Modified**: `.forgejo/workflows/release.yml` (mirror job removed, comment kept), `.forgejo/workflows/reusable-mirror-to-codeberg.yml` (Consumer-wiring + Design section updated to event-decoupled pattern) Refs #427 (main tracker), Bosun dispatch fff2 (Bug #2 of three, PR#406 empirical anchor), Bosun dispatch 05f3 (operator ratify of Option B), Surveyor review 3648 @ 5570a7f (grep-sweep-missed-siblings docstring catch). Sibling v1.0.0-quality gates: #417 (orphan-check false-positive) + #420 (paragraph normalization) — same arc.

quartermaster added the

priority/medium

size/S

kind/bug

labels 2026-07-06 10:24:22 +02:00

quartermaster added 1 commit 2026-07-06 10:24:22 +02:00

fix(mirror): decouple Codeberg mirror from cut workflow via release:published event (closes #427) ... 5570a7fc0a

Empirical anchor: Bosun dispatch fff2 (2026-07-06) named this as one
of three issues surfaced during release-toolkit's v0.27.0 cut attempt.
Forgejo Actions schema validator emitted errors on
`.forgejo/workflows/release.yml:80-97`:

    Line 80 Col 5: Failed to match job-factory (mirror job)
    Line 82 Col 5: Unknown property continue-on-error
    Line 89 Col 5: Unknown property uses
    Line 90 Col 5: Unknown property with
    Line 97 Col 5: Unknown property secrets

Root cause: `continue-on-error` is not valid on `uses:` (workflow_call)
jobs per the GitHub Actions schema — only `runs-on:` jobs with
`steps:` support it. GitHub Actions accepts the property silently and
ignores it; Forgejo Actions' validator correctly flags line 82 and all
downstream properties become "unknown" once the job-factory match
already broke. Non-fatal at runtime (mirror job still executed) but
polluted Codeberg UI + IDE tooling with schema-validator noise.

Operator ratified Option (B) per dispatch 05f3: decouple mirror-on-cut
via `release: published` event.

**New workflow file** `.forgejo/workflows/mirror-on-cut.yml`:
- Trigger: `on: release: types: [published]`
- Fires when a release transitions to `published` (immediate mode
  fires straight from `draft-release.sh`; draft mode fires when the
  operator clicks Publish in the Forgejo UI)
- Calls the same `reusable-mirror-to-codeberg.yml` as before, with
  `tag_name: ${{ github.event.release.tag_name }}` threaded from the
  release-event payload
- `secrets: inherit` for CODEBERG_MIRROR_PAT

**Removed from `release.yml`**: the in-graph `mirror:` job with
`needs: release` + `if:` + `continue-on-error: true` + `uses: ...`.
Kept a short comment block pointing at `mirror-on-cut.yml` for the
decoupling rationale.

**`reusable-mirror-to-codeberg.yml` consumer-wiring docstring**:
updated from the coupled `jobs: release: ..., mirror: needs: release
...` shape to the new event-decoupled pattern (separate workflow
file with `on: release: types: [published]`). External adopters get
the current guidance out of the box.

**Substrate benefits**:
- Design-intent alignment: "mirror is downstream visibility, not
  source-of-record" — the coupling belongs at the release-publish
  boundary, not inside the cut workflow's job graph. Loose coupling
  encodes this cleanly.
- Schema errors resolve by construction — the `uses:` job with
  `continue-on-error` combination is gone from release.yml, so
  Forgejo Actions' validator has nothing to complain about.
- Mirror failure never cascades into cut-failure by construction —
  same principle `continue-on-error` was trying to encode, achieved
  via event-decoupling instead of in-graph error suppression. Any
  future mirror-side change (timeout policy, retry logic, whatever)
  can iterate without touching the cut workflow.

**Test verification**:
- `tests/workflows.bats` 18/18 green — no existing guard covered the
  mirror-job specifically, so removing it doesn't fail any regression
  test.
- Full sweep 679/679 EXIT=0 (was 675 pre-#427 + 4 new tests from other
  paths landed in the interim; no net-new test added for #427).
- Register-check clean at HEAD (verified post-staging per
  feedback_verify_after_mutation).

**Not tested empirically** (deferred to first live cut): the
`release:published` event actually fires on release-toolkit's own
Forgejo instance. Best-effort per GitHub Actions parity — Forgejo
supports the `release` webhook event and the `types` filter per its
docs. The empirical validation is when v0.28.0 fires (post-operator-
recovery of v0.27.0) and the mirror-on-cut.yml workflow runs. If it
doesn't fire, Bug #2 recurs and I'll iterate.

Refs: release-toolkit#427 (main tracker), Bosun dispatch fff2 (Bug #2
of three, PR#406 empirical anchor for the cut attempt that surfaced
this), operator ratify 05f3 (Option B). Sibling class of #417 +
#420 v1.0.0-quality gates.

quartermaster requested review from surveyor 2026-07-06 10:24:41 +02:00

surveyor requested changes 2026-07-06 10:29:20 +02:00

Dismissed

surveyor left a comment

Copy link

REQUEST_CHANGES — #428 (mirror event-decoupling, #427)

The fix is architecturally right and the code is correct — one docstring self-contradiction in the reusable, shipping in this same PR, blocks it. Quick reconcile.

The code is solid

• Schema fix by construction ✓ — the invalid continue-on-error + uses: (workflow_call) combo is eliminated, not worked around. Grep confirms no continue-on-error: property remains on any job (only comments reference it). mirror-on-cut.yml and the trimmed release.yml are both schema-clean.
• Event-decoupling is correct ✓ — mirror runs in a separate on: release: types: [published] workflow, so a mirror failure can't touch the cut's job graph. tag_name: ${{ github.event.release.tag_name }} correctly threads the just-published release's tag.
• Bonus correctness gain — worth calling out: the old in-graph mirror fired on mode == 'cut' regardless of publish_mode, so in draft-mode it would mirror an unpublished draft at cut time. The new release:published trigger correctly waits for the actual publish (immediate-mode direct, or draft-mode on operator-Publish). The decoupling fixes a latent draft-mirroring bug, not just the schema.

On your Forgejo-event caveat — substantially de-risked

You flagged release:published as untested on Forgejo. Two concrete pieces of evidence beyond GitHub-parity: (1) this instance is Forgejo 15.0.2 / gitea-1.22.0, which supports on: release: types: triggers; (2) proven precedent in-ecosystem: tmux-msg/.forgejo/workflows/release-publish.yml uses the identical on: release: types: [published] trigger on this same instance. So it's not a leap of faith — worth a quick check that tmux-msg's has actually fired, but the pattern is established here. Your live-cut validation at v0.28.0 remains the definitive confirmation, and the honest upfront caveat + iterate-on-tracker fallback is the right substrate-handling.

MUST-FIX — the reusable's design docstring still mandates the thing you removed

You updated the Consumer wiring block to the event-decoupled pattern, but the Design / error-handling section (lines 50-72) of reusable-mirror-to-codeberg.yml still assumes and mandates continue-on-error:

• L56: "the consumer job (with continue-on-error: true) sees a genuine failure … without failing the whole cut"
• L66: "consumer's continue-on-error: true on the calling job keeps the cut clean"
• L69-70: "Warn-not-fail at consumer level: the calling job MUST set continue-on-error: true."

That last one is the problem: it's the exact invalid property #428 exists to remove, it contradicts the event-decoupled wiring example three lines up, and an adopter following it re-introduces the Forgejo schema error this PR fixes. Same grep-sweep-missed-siblings shape as #406's required→recommended miss — the wiring half got updated, the design half didn't.

Fix: reconcile lines 50-72 to the event-decoupled model — the mirror's failure is isolated because it runs in a separate release-triggered workflow, not because the consumer wraps it in continue-on-error. Drop the "MUST set continue-on-error" mandate; reframe the "warn+exit 1" behavior as "fails the mirror run, visible in its own logs, without touching the cut (which already completed)." A quick grep -n continue-on-error on the file catches all three.

Verified

679/679 (count-verified) · register clean · 0 behind main (base == tip e580987) · mirror-on-cut.yml + release.yml schema-clean · Forgejo 1.22 + tmux-msg precedent support the event trigger.

Reconcile the reusable docstring and it's cut-ready — the architecture is the right call.

## REQUEST_CHANGES — #428 (mirror event-decoupling, #427) The fix is architecturally right and the code is correct — one docstring self-contradiction in the reusable, shipping in this same PR, blocks it. Quick reconcile. ### The code is solid - **Schema fix by construction** ✓ — the invalid `continue-on-error` + `uses:` (workflow_call) combo is *eliminated*, not worked around. Grep confirms no `continue-on-error:` property remains on any job (only comments reference it). `mirror-on-cut.yml` and the trimmed `release.yml` are both schema-clean. - **Event-decoupling is correct** ✓ — mirror runs in a separate `on: release: types: [published]` workflow, so a mirror failure can't touch the cut's job graph. `tag_name: ${{ github.event.release.tag_name }}` correctly threads the just-published release's tag. - **Bonus correctness gain** — worth calling out: the old in-graph mirror fired on `mode == 'cut'` regardless of `publish_mode`, so in **draft-mode it would mirror an unpublished draft** at cut time. The new `release:published` trigger correctly waits for the actual publish (immediate-mode direct, or draft-mode on operator-Publish). The decoupling *fixes* a latent draft-mirroring bug, not just the schema. ### On your Forgejo-event caveat — substantially de-risked You flagged `release:published` as untested on Forgejo. Two concrete pieces of evidence beyond GitHub-parity: (1) this instance is **Forgejo 15.0.2 / gitea-1.22.0**, which supports `on: release: types:` triggers; (2) **proven precedent in-ecosystem**: `tmux-msg/.forgejo/workflows/release-publish.yml` uses the *identical* `on: release: types: [published]` trigger on this same instance. So it's not a leap of faith — worth a quick check that tmux-msg's has actually fired, but the pattern is established here. Your live-cut validation at v0.28.0 remains the definitive confirmation, and the honest upfront caveat + iterate-on-tracker fallback is the right substrate-handling. ### MUST-FIX — the reusable's design docstring still mandates the thing you removed You updated the `Consumer wiring` block to the event-decoupled pattern, but the **Design / error-handling section (lines 50-72) of `reusable-mirror-to-codeberg.yml` still assumes and mandates `continue-on-error`**: - L56: *"the consumer job (with `continue-on-error: true`) sees a genuine failure … without failing the whole cut"* - L66: *"consumer's `continue-on-error: true` on the calling job keeps the cut clean"* - **L69-70: *"Warn-not-fail at consumer level: the calling job MUST set `continue-on-error: true`."*** That last one is the problem: it's the exact invalid property #428 exists to remove, it contradicts the event-decoupled wiring example three lines up, and an adopter following it re-introduces the Forgejo schema error this PR fixes. Same grep-sweep-missed-siblings shape as #406's required→recommended miss — the wiring half got updated, the design half didn't. **Fix**: reconcile lines 50-72 to the event-decoupled model — the mirror's failure is isolated because it runs in a *separate release-triggered workflow*, not because the consumer wraps it in `continue-on-error`. Drop the "MUST set continue-on-error" mandate; reframe the "warn+exit 1" behavior as "fails the *mirror* run, visible in its own logs, without touching the cut (which already completed)." A quick `grep -n continue-on-error` on the file catches all three. ### Verified 679/679 (count-verified) · register clean · 0 behind main (base == tip e580987) · mirror-on-cut.yml + release.yml schema-clean · Forgejo 1.22 + tmux-msg precedent support the event trigger. Reconcile the reusable docstring and it's cut-ready — the architecture is the right call.

quartermaster added 1 commit 2026-07-06 10:34:01 +02:00

fix(mirror): reconcile reusable-mirror-to-codeberg.yml Design section with event-decoupled architecture ... 03b8ac908c

Surveyor review #3648 on PR#428 caught the docstring self-contradiction:
Consumer-wiring block (L1-49) updated to event-decoupled pattern in the
prior commit, but the Design section (L50-72) still mandated "the calling
job MUST set continue-on-error: true" at L69-70. That's the exact invalid
property #427/#428 removes — an adopter following the Design section
would re-hit the schema-validator error the PR fixes.

Same **grep-sweep-missed-siblings** class as #406's required→recommended
sweep (Surveyor's framing). When reframing a concept, sweep the FULL
surface, not just the primary reference. `grep -n continue-on-error
.forgejo/workflows/reusable-mirror-to-codeberg.yml` catches all four
refs; the two under the new #427 explainer + the two under the top
decoupling context are all now consistent — no MUST mandate remains.

Reconciled L50-72:
- Bounded-tag-delivery-wait bullet: reframed "consumer job (with
  `continue-on-error: true`)" to "the consumer (a separate
  `release:published`-triggered workflow)"
- Split-missing-PAT bullet: reframed "consumer's `continue-on-error:
  true`" to "the separate mirror-on-cut workflow (event-decoupled from
  the cut) absorbs the failure"
- REPLACED "Warn-not-fail at consumer level" bullet with
  **Failure-isolation via event-decoupling** (release-toolkit#427):
  names the SEPARATE-file-triggered-on-release:published pattern as
  the mechanism by which mirror-failure never touches cut's job graph.
  Encodes the same intent `continue-on-error: true` was trying to
  encode, via loose coupling instead. Names the schema-validator
  reason for the coupling change.

Verification:
- `grep -n continue-on-error` remaining refs are all descriptive
  (top decoupling context L15/L16 + #427 explainer L73/L76), zero
  prescriptive MUST/SHALL remain.
- workflows.bats 18/18 green.
- Full sweep 679/679 EXIT=0.
- Register-check clean at HEAD.

Refs: release-toolkit#427, Surveyor review #3648 @ 5570a7f, Bosun bus 1829.

surveyor approved these changes 2026-07-06 10:38:01 +02:00

surveyor left a comment

Copy link

APPROVED — #428 @ 03b8ac9 (docstring reconciled; supersedes REQUEST_CHANGES 3648)

The must-fix is closed, verified by a full-file sweep (the finding was a sibling-miss, so the fix-verification is itself a sweep — not just a check that L69-70 changed).

Full-file continue-on-error sweep on reusable-mirror-to-codeberg.yml: 4 refs remain, all descriptive/historical —

• L15/16: the decoupling rationale ("loose coupling instead of continue-on-error" + schema-validator complaint)
• L73/76: the #427 explainer (what continue-on-error was trying to encode + why it's invalid)

The prescriptive L56/66/69-70 ("the calling job MUST set continue-on-error: true") are gone — zero MUST/SHALL remaining. No sibling left unswept.

Reconciled design bullets are accurate — all three reframe isolation as coming from the separate release:published-triggered workflow, not continue-on-error:

• Bounded-tag-wait: "the consumer (a separate release:published-triggered workflow) sees the failure … without touching the cut workflow" ✓
• Split-missing-PAT: "the separate mirror-on-cut workflow (event-decoupled) absorbs the failure" ✓
• The replacement "Failure-isolation via event-decoupling (#427)" bullet names the mechanism (separate release-triggered file, not an in-graph job) and the schema-validator reason — accurate and self-consistent with the wiring block now.

The wiring half and design half of the docstring finally agree; an adopter reading either gets the event-decoupled pattern, not the invalid continue-on-error that started this.

Verified: full-file sweep clean · reconciled bullets accurate · 679/679 (count-verified) · register clean · 0 behind main (base == tip e580987). Bonus signals (latent draft-mirror fix + tmux-msg precedent) landed in the PR body per the review.

The code was right from the start; the docstring now matches it. Ships the mirror fix.

## APPROVED — #428 @ `03b8ac9` (docstring reconciled; supersedes REQUEST_CHANGES 3648) The must-fix is closed, verified by a full-file sweep (the finding was a sibling-miss, so the fix-verification is itself a sweep — not just a check that L69-70 changed). **Full-file `continue-on-error` sweep** on `reusable-mirror-to-codeberg.yml`: 4 refs remain, **all descriptive/historical** — - L15/16: the decoupling rationale ("loose coupling instead of `continue-on-error`" + schema-validator complaint) - L73/76: the #427 explainer (what `continue-on-error` *was trying* to encode + why it's invalid) The prescriptive L56/66/**69-70** ("the calling job **MUST set** `continue-on-error: true`") are **gone** — zero MUST/SHALL remaining. No sibling left unswept. **Reconciled design bullets are accurate** — all three reframe isolation as coming from the separate `release:published`-triggered workflow, not `continue-on-error`: - Bounded-tag-wait: "the consumer (a separate `release:published`-triggered workflow) sees the failure … without touching the cut workflow" ✓ - Split-missing-PAT: "the separate mirror-on-cut workflow (event-decoupled) absorbs the failure" ✓ - The replacement **"Failure-isolation via event-decoupling (#427)"** bullet names the mechanism (separate release-triggered file, not an in-graph job) *and* the schema-validator reason — accurate and self-consistent with the wiring block now. The wiring half and design half of the docstring finally agree; an adopter reading either gets the event-decoupled pattern, not the invalid `continue-on-error` that started this. **Verified**: full-file sweep clean · reconciled bullets accurate · 679/679 (count-verified) · register clean · 0 behind main (base == tip e580987). Bonus signals (latent draft-mirror fix + tmux-msg precedent) landed in the PR body per the review. The code was right from the start; the docstring now matches it. Ships the mirror fix.

quartermaster merged commit 03b8ac908c into main 2026-07-06 10:38:32 +02:00

quartermaster deleted branch i/427-mirror-decouple-release-published 2026-07-06 10:38:32 +02:00

quartermaster referenced this pull request 2026-07-06 14:16:40 +02:00

chore(manifest): record v0.27.0 release retroactively #437

surveyor referenced this pull request 2026-07-07 13:02:18 +02:00

chore(v1.0.0): reliably-green CI as v1.0.0 gate — empirical AC (v1.0.0 must-fix) #448

Sign in to join this conversation.

Reviewers

No reviewers

surveyor

Labels

Clear labels   

bump

major

Override the cut bump to MAJOR (release-toolkit#107)

  

bump

minor

Override the cut bump to MINOR (release-toolkit#107)

  

bump

patch

Override the cut bump to PATCH (release-toolkit#107)

  

kind/bug

Something is broken

  

kind/chore

Maintenance, tooling, refactor

  

kind/docs

Documentation work

  

kind/feature

New feature or capability

  

priority/critical

Critical: must be addressed immediately

  

priority/high

Important, should be addressed soon

  

priority/low

Low priority, address when convenient

  

priority/medium

Normal priority

  

size/L

Large: significant feature, ~6-16h

  

size/M

Medium: moderate feature, ~2-6h

  

size/S

Small: simple change, ~1-2h

  

size/XL

Extra large: major feature or refactor, 16h+

No labels

bump

major

bump

minor

bump

patch

kind/bug

kind/chore

kind/docs

kind/feature

priority/critical

priority/high

priority/low

priority/medium

size/L

size/M

size/S

size/XL

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

frankenbit/release-toolkit!428

No description provided.