Artifact [8a9349ab53]

Artifact 8a9349ab53f6eeb1aea4af9f5279b2a4c8bd1d9770dad8c2ef0306a137f223e3:

• File docs/superpowers/specs/2026-04-30-cross-repo-release-automation-design.md — part of check-in [6273a0750d] at 2026-07-22 23:03:24 on branch trunk — fix: correct go.mod module path to go-libfossil (#76) (#117)

  * fix: correct go.mod module path to go-libfossil (#76)

  The module directive still declared github.com/danmestas/libfossil after the repository was renamed to go-libfossil. pkg.go.dev/godoc scraping matches the declared module path against its hosting path, so the stale path broke SDK doc generation.

  Rename the module path across all four in-repo modules (root, the two db/driver submodules, observer/otel, and docs/site) and every internal import, plus the release workflow, the doc-generation script, and the docs.

  No tag has been published under the go-libfossil path yet, so the cross-module requires cannot resolve from the proxy. Add replace directives pointing the in-repo requires at their local directories so GOWORK=off builds and the CI matrix resolve without a published version. These replaces are ignored by downstream consumers and can be dropped once new-path tags are cut.

  * Guard releases against replace directives; regenerate SDK docs

  The go-libfossil rename bridges the in-repo modules with `replace` directives. Go refuses a main module that carries them, so a tag cut while they are present breaks `go install .../cmd/libfossil@<tag>` and every `@version` fetch of the driver submodules. Add a Release workflow step that fails the job before anything is published if any tracked go.mod still contains a replace directive, so the bridge is a CI-enforced TODO rather than a silent landmine. The step comment records what unblocks a release: re-tag the submodules with the corrected module paths, bump the root requires, then drop the replace blocks.

  Regenerate the SDK reference docs. The previous pass only substituted URLs over the pre-existing generated files, leaving stale signatures (Repo.Timeline was still documented as taking LogOpts rather than TimelineOpts). Note that gomarkdoc must be run outside a git worktree — inside one it silently strips every source link.

  Also reword the README rename note, which read as though the import path had not changed, and re-sort the ncruces require block. (user: [email protected], size: 25901) [annotate] [blame] [check-ins using]

# Cross-repo release automation — design spec

- **Date**: 2026-04-30
- **Status**: Draft, pending implementation
- **Scope**: libfossil, edgesync, bones — three Go modules with a dependency chain
- **Author**: Dan Mestas

## Background

Three Go modules form a dependency chain:

```
libfossil ──> edgesync ──> bones
```

bones also imports libfossil directly (a diamond), but this design treats the chain as linear and accepts that bones receives libfossil updates only when edgesync re-releases. See "Decisions log" for rationale.

The most recent cascade (libfossil v0.4.5 → edgesync v0.0.7 → bones v0.3.3, all 2026-04-30) exposed the manual cost of the chain:

- libfossil tagged v0.4.5 with a sync bug fix.
- ~3 days passed before edgesync was bumped + tagged v0.0.7 (no notification mechanism).
- bones then required two separate PRs (#85, #86) to bump libfossil and edgesync — one because the dependency graph fans in, and one because there's no consolidation logic.
- Each repo's tag annotation was inconsistent: bones v0.3.3 had hand-written multi-line release notes; libfossil/edgesync had bare tags.
- A pre-existing rule in project memory ("run CI locally every time after you pr") was not catching CI-only failures because there's no enforcement and no single command that mirrors what GitHub Actions runs.

Pre-existing CI backlog (deferred 2026-04-26, see `project_pending_ci_improvements.md`): `paths-ignore` for docs PRs, `-race`, SDK-drift check, DST seed sweep. Not in scope here; tracked separately.

## Goals

1. Reduce upstream-tag → downstream-PR latency from days to minutes.
2. Eliminate the "two PRs to bones per cascade" failure mode.
3. Replace the discipline-based "run CI locally" rule with a fast, automatic pre-push check that catches obvious breakage.
4. Provide one command (`make ci`) per repo that runs the *exact* CI workflow, so contributors and the maintainer can verify before pushing.
5. Standardize release notes across all three repos (consistent quality, less manual annotation work).
6. Provide a `make release VERSION=...` helper that validates and tags from a template.

## Non-goals

- Auto-merging PRs. Maintainer reviews every PR per `~/.claude/CLAUDE.md` policy.
- Renovate/Dependabot. The repository_dispatch chain replaces it; Dependabot's polling adds latency without benefit for first-party deps.
- Monorepo migration.
- Cross-repo CI (running edgesync's tests on a libfossil PR). Out of scope; possible follow-up.
- The deferred CI backlog (paths-ignore, race, drift, DST sweep) — separate PR.

## Architecture overview

End-state of the cascade:

```
   libfossil tag (e.g. v0.4.6)
        │
        │ release.yml fires:
        │   - gh release create --generate-notes
        │   - gh api repos/danmestas/EdgeSync/dispatches  (event_type=upstream-bump)
        ▼
     edgesync
        │ bump-upstream.yml fires:
        │   - go get [email protected] && go mod tidy && go build ./...
        │   - opens PR "chore: bump libfossil → vX.Y.Z" on branch chore/bump-libfossil-vX.Y.Z
        │
        │ maintainer reviews, CI passes, merges
        │ maintainer runs: make release VERSION=v0.0.8
        │
        │ release.yml fires:
        │   - goreleaser publishes GitHub release with auto-changelog
        │   - gh api bones/dispatches
        ▼
       bones
        │ bump-upstream.yml fires:
        │   - go get [email protected] && go mod tidy
        │   - libfossil bump comes along automatically via Go MVS (edgesync's go.sum pins it)
        │   - opens single PR "chore: bump EdgeSync → vA.B.C"
        │
        │ maintainer reviews, merges, tags, ships goreleaser → Homebrew
```

Each repo additionally gains:

- `make ci` — verbatim mirror of that repo's GitHub Actions test workflow.
- `make ci-fast` — vet + build + short tests, sub-60s, used by pre-push hook.
- `.git/hooks/pre-push` (auto-installed via `make setup-hooks`) — runs `make ci-fast`, blocks push on failure.
- `make release VERSION=vX.Y.Z` — validates clean tree, runs `make ci`, opens `$EDITOR` with auto-populated tag annotation template, creates annotated tag locally. Push is intentionally manual.
- `.github/RELEASE_TEMPLATE.md` — boilerplate filled in by `make release`.
- GoReleaser changelog config (edgesync, bones) — auto-grouped release notes.

## Per-repo components

### libfossil (top of chain — currently has no release pipeline)

> **Deviation from PR #1/#2 (decided 2026-04-30, option A):** libfossil's existing `.githooks/pre-commit` is comprehensive (~45s — modernc + ncruces + vet + otel + build + SDK drift) and exceeds what `make ci-fast` would do. Therefore PR #3 does NOT add a `pre-push` hook or a `ci-fast` target. The existing pre-commit is libfossil's sole local gate; `make ci` is for manual full-CI verification before tagging.

**New files**:
- `.github/workflows/release.yml` (PR #6) — triggers on `push: tags: [v*]`. Two steps: `gh release create $TAG --generate-notes --verify-tag` and the dispatch step to edgesync.
- `Makefile` targets (PR #3, merged): `ci`, `ci-default`, `ci-ncruces`, `ci-otel-target`, `release`. (No `ci-fast` per the deviation above.)
- `.github/RELEASE_TEMPLATE.md` (PR #3, merged).

**Existing — unchanged**: `.github/workflows/test.yml`, `.githooks/pre-commit`, `setup-hooks` Makefile target, `test/test-drivers/test-otel` Makefile targets (the dev-friendly fast-path; intentionally distinct from the new `ci-*` targets which mirror CI verbatim).

**No goreleaser**: libfossil is a library — no binaries are published. `gh release create --generate-notes` provides auto-changelog and source tarballs without 50 lines of goreleaser config. Revisit if `cmd/libfossil` ever ships as a published binary.

**Multi-module note**: libfossil has 3 sub-modules with their own `go.mod` (`db/driver/modernc/`, `db/driver/ncruces/`, `observer/otel/`). The new `ci-default`/`ci-ncruces`/`ci-otel-target` targets handle this via explicit `cd <subdir> && go test ./...` steps. If a future `ci-fast` is ever added, it must do the same — `go test ./...` from repo root would silently skip the sub-modules.

**Secret needed**: `DISPATCH_PAT` — fine-grained PAT, repo access scoped to `danmestas/EdgeSync` only, permission `Contents: Read and write` (for PR #6).

### edgesync (middle — receives + dispatches)

> **Naming note**: the GitHub repo is `danmestas/EdgeSync` (mixed case, matches `go.mod` import path `github.com/danmestas/EdgeSync`). All `gh api repos/danmestas/EdgeSync/...` calls and `${{ github.event.repository.name }}` outputs use this canonical case. Local clone path `~/projects/edgesync` is lowercase but irrelevant to the workflow.

**New files**:
- `.github/workflows/bump-upstream.yml` — receiver listening for `repository_dispatch: types: [upstream-bump]` plus `workflow_dispatch` manual fallback.
- Same Makefile + hook + template files as libfossil.

**Modified files**:
- `.github/workflows/release.yml` — append a dispatch step targeting bones.
- `.goreleaser.yml` — add `changelog: { use: github, sort: asc, groups: [...] }` config.

**Secret needed**: `DISPATCH_PAT` — fine-grained PAT, repo access scoped to `danmestas/bones` only, `Contents: Read and write`.

**EXPECTED_REPO** in receiver workflow: `libfossil` (matches `${{ github.event.repository.name }}` from libfossil's release.yml). Receiver no-ops on any other dispatch.

### bones (leaf — receives only)

**New files**:
- `.github/workflows/bump-upstream.yml` — receiver only, no sender.
- Same Makefile + hook + template files.

**Modified files**:
- `.goreleaser.yml` — add changelog config.
- `.github/workflows/release.yml` — no changes (bones doesn't dispatch).

**Secret needed**: none beyond existing.

**EXPECTED_REPO** in receiver workflow: `EdgeSync`.

## Dispatch protocol

**Event name**: `upstream-bump` (single event, receiver discriminates on payload).

**Payload schema** (`client_payload`):

| Field | Example | Purpose |
|-------|---------|---------|
| `repo` | `libfossil` | Short name; receiver checks against `EXPECTED_REPO` env var |
| `module_path` | `github.com/danmestas/go-libfossil` | Full Go module path used by `go get` |
| `version` | `v0.4.6` | Tag string with leading `v`; validated against `^v\d+\.\d+\.\d+(-.+)?$` |
| `release_url` | `https://github.com/danmestas/go-libfossil/releases/tag/v0.4.6` | Embedded in PR body for one-click reviewer access |
| `tag_sha` | `2baaced...` | Audit trail in case of force-moved tags |
| `actor` | `danmestas` | Who pushed the tag |

**Validation**: receiver workflow fails loud (`::error::`) on:
- `repo` not matching `EXPECTED_REPO` → exits 0 (silent ignore for cross-routing safety).
- `module_path` not matching `EXPECTED_MODULE` → exits 1.
- `version` failing regex → exits 1.

**Sender step pattern** (in libfossil/edgesync `release.yml`):

```yaml
- name: Dispatch downstream bump
  env:
    GH_TOKEN: ${{ secrets.DISPATCH_PAT }}
    DOWNSTREAM: EdgeSync                            # in libfossil; "bones" in edgesync
    MODULE_PATH: github.com/danmestas/go-libfossil    # the upstream's own module path; in edgesync: github.com/danmestas/EdgeSync
  run: |
    gh api -X POST "repos/danmestas/${DOWNSTREAM}/dispatches" \
      -f event_type=upstream-bump \
      -f client_payload[repo]=${{ github.event.repository.name }} \
      -f client_payload[module_path]=${MODULE_PATH} \
      -f client_payload[version]=${{ github.ref_name }} \
      -f client_payload[release_url]=${{ github.server_url }}/${{ github.repository }}/releases/tag/${{ github.ref_name }} \
      -f client_payload[tag_sha]=${{ github.sha }} \
      -f client_payload[actor]=${{ github.actor }}
```

`DOWNSTREAM` and `MODULE_PATH` are hard-coded per repo:
- libfossil's release.yml: `DOWNSTREAM=EdgeSync`, `MODULE_PATH=github.com/danmestas/go-libfossil`
- edgesync's release.yml: `DOWNSTREAM=bones`, `MODULE_PATH=github.com/danmestas/EdgeSync`

**Idempotency**: Receiver checks for an existing open PR on branch `chore/bump-${repo}-${version}` before doing work. Re-firing the same dispatch is a no-op.

**Failure recovery**:

| Failure | Symptom | Recovery |
|---------|---------|----------|
| `DISPATCH_PAT` expired | Sender's dispatch step fails; tag + release still created | Rotate PAT, manually fire via `gh workflow run bump-upstream.yml` |
| Receiver workflow broken | No PR opens after upstream tag | Manually bump go.mod, fix workflow on a separate PR |
| `go get` fails (module proxy lag) | Receiver auto-retries 3× with 15s backoff; if still failing, workflow fails in "Bump module" step | Re-fire via `workflow_dispatch` after a few minutes (most cases self-recover) |
| Bad tag pushed (force-move) | `tag_sha` in PR body shows the old SHA | Reviewer notices in PR body before merging |

## Receiver workflow (`bump-upstream.yml`)

Same shape in edgesync and bones. ~75 lines including `workflow_dispatch` fallback.

```yaml
name: Bump upstream
on:
  repository_dispatch:
    types: [upstream-bump]
  workflow_dispatch:
    inputs:
      repo:        { required: true,  type: string }
      module_path: { required: true,  type: string }
      version:     { required: true,  type: string }
      release_url: { required: false, type: string }
      tag_sha:     { required: false, type: string }
      actor:       { required: false, type: string }

permissions:
  contents: write
  pull-requests: write

jobs:
  bump:
    runs-on: ubuntu-latest
    env:
      EXPECTED_REPO: libfossil          # bones uses "EdgeSync"
      EXPECTED_MODULE: github.com/danmestas/go-libfossil  # bones: github.com/danmestas/EdgeSync
    steps:
      - name: Resolve inputs
        id: in
        run: |
          REPO="${{ github.event.client_payload.repo || inputs.repo }}"
          MODULE_PATH="${{ github.event.client_payload.module_path || inputs.module_path }}"
          VERSION="${{ github.event.client_payload.version || inputs.version }}"
          RELEASE_URL="${{ github.event.client_payload.release_url || inputs.release_url }}"
          TAG_SHA="${{ github.event.client_payload.tag_sha || inputs.tag_sha }}"
          if [[ "$REPO" != "$EXPECTED_REPO" ]]; then
            echo "::notice::Ignoring dispatch for $REPO (expected $EXPECTED_REPO)"
            exit 0
          fi
          [[ "$MODULE_PATH" == "$EXPECTED_MODULE" ]] || { echo "::error::module_path mismatch: got $MODULE_PATH, expected $EXPECTED_MODULE"; exit 1; }
          [[ "$VERSION" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-.+)?$ ]] || { echo "::error::bad version: $VERSION"; exit 1; }
          {
            echo "repo=$REPO"
            echo "module=$MODULE_PATH"
            echo "version=$VERSION"
            echo "release_url=$RELEASE_URL"
            echo "tag_sha=$TAG_SHA"
            echo "branch=chore/bump-${REPO}-${VERSION}"
          } >> "$GITHUB_OUTPUT"

      - name: Idempotency check
        id: idem
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          if gh pr list --repo "${{ github.repository }}" --head "${{ steps.in.outputs.branch }}" --state open --json number -q '.[0].number' | grep -q .; then
            echo "exists=true" >> "$GITHUB_OUTPUT"
            echo "::notice::PR already open for ${{ steps.in.outputs.branch }}; skipping"
          fi

      - if: steps.idem.outputs.exists != 'true'
        uses: actions/checkout@v4
        with: { fetch-depth: 0 }

      - if: steps.idem.outputs.exists != 'true'
        uses: actions/setup-go@v5
        with: { go-version-file: go.mod }

      - if: steps.idem.outputs.exists != 'true'
        name: Bump module
        id: bump
        run: |
          git switch -c "${{ steps.in.outputs.branch }}"
          # GOPROXY can lag the upstream tag push; retry before giving up.
          for i in 1 2 3; do
            if go get "${{ steps.in.outputs.module }}@${{ steps.in.outputs.version }}"; then
              break
            fi
            [ "$i" -lt 3 ] && sleep 15 || { echo "::error::go get failed after 3 attempts"; exit 1; }
          done
          go mod tidy
          if git diff --quiet go.mod go.sum; then
            echo "::notice::No-op bump — version already pinned"
            echo "noop=true" >> "$GITHUB_OUTPUT"
          fi

      - if: steps.idem.outputs.exists != 'true' && steps.bump.outputs.noop != 'true'
        name: Verify build
        run: go build ./...

      - if: steps.idem.outputs.exists != 'true' && steps.bump.outputs.noop != 'true'
        name: Commit + push + open PR
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          git config user.name  "github-actions[bot]"
          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
          git add go.mod go.sum
          git commit -m "chore: bump ${{ steps.in.outputs.repo }} → ${{ steps.in.outputs.version }}"
          git push -u origin "${{ steps.in.outputs.branch }}"
          gh pr create \
            --title "chore: bump ${{ steps.in.outputs.repo }} → ${{ steps.in.outputs.version }}" \
            --body "Auto-bump from upstream release.

          - Module: \`${{ steps.in.outputs.module }}\`
          - Version: \`${{ steps.in.outputs.version }}\`
          - Upstream release: ${{ steps.in.outputs.release_url }}
          - Tag SHA: \`${{ steps.in.outputs.tag_sha }}\`

          CI must pass before merge. Review go.sum diff for transitive surprises."
```

## Local dev surface

### `make ci` — per-repo verbatim mirrors

**libfossil** (mirrors `test.yml`):

```make
.PHONY: ci ci-default ci-ncruces ci-otel ci-fast

ci: ci-default ci-ncruces ci-otel

ci-default:
	GOWORK=off go test $(shell GOWORK=off go list ./... | grep -v '/dst') -count=1 -timeout=120s
	GOWORK=off go test ./dst/... -count=1 -timeout=300s
	cd db/driver/modernc && GOWORK=off go test ./... -count=1
	GOWORK=off go vet ./...
	GOWORK=off go build ./cmd/libfossil/

ci-ncruces:
	GOWORK=off go test -tags test_ncruces $(shell GOWORK=off go list ./... | grep -v '/dst' | grep -v 'cmd/libfossil') -count=1 -timeout=120s
	GOWORK=off go test -tags test_ncruces ./dst/... -count=1 -timeout=300s
	cd db/driver/ncruces && GOWORK=off go test ./... -count=1

ci-otel:
	cd observer/otel && GOWORK=off go test ./... -count=1
```

**edgesync** (mirrors `ci.yml` + `test.yml`): targets `ci-vet`, `ci-leaf-bridge`, `ci-cmd`, `ci-sim`. Exact commands extracted verbatim from each step's `run:` block.

**bones** (mirrors the bones-side portion of `ci.yml`; cross-repo leaf-binary integration is CI-only):

```make
ci:
	$(MAKE) check
	go build -tags=otel ./...
	go test -tags=otel -short ./... -count=1
```

### `make ci-fast` — pre-push subset

```make
ci-fast:
	go vet ./...
	go build ./...
	go test -short -count=1 -timeout=30s ./...
```

No `-race`, no `-tags=otel`, no `-tags test_ncruces`, no driver subdirs. ~30-60s. Catches build break, vet errors, obvious test failures. Misses: race conditions, otel-tagged code, ncruces-tagged code, DST. Those are `make ci` territory.

### `scripts/install-hooks.sh`

```sh
#!/usr/bin/env bash
set -euo pipefail
HOOK_DIR="$(git rev-parse --git-dir)/hooks"
mkdir -p "$HOOK_DIR"
cat > "$HOOK_DIR/pre-push" <<'EOF'
#!/usr/bin/env bash
set -e
echo "[pre-push] running make ci-fast..."
make ci-fast
EOF
chmod +x "$HOOK_DIR/pre-push"
echo "Installed pre-push hook → make ci-fast"
```

Wired into existing `make setup-hooks`. Bypass with `git push --no-verify` only when intentional.

### `make release VERSION=...`

```make
.PHONY: release
release:
	@test -n "$(VERSION)" || { echo "VERSION=vX.Y.Z required"; exit 1; }
	@echo "$(VERSION)" | grep -qE '^v[0-9]+\.[0-9]+\.[0-9]+(-.+)?$$' || { echo "bad version format"; exit 1; }
	@git diff --quiet || { echo "tree dirty; commit or stash first"; exit 1; }
	@git fetch origin --tags
	@if git rev-parse "$(VERSION)" >/dev/null 2>&1; then echo "tag $(VERSION) already exists"; exit 1; fi
	@$(MAKE) ci
	@PREV=$$(git describe --tags --abbrev=0 2>/dev/null || echo ""); \
	  TMPL=.github/RELEASE_TEMPLATE.md; \
	  TMP=$$(mktemp); \
	  { echo "Release $(VERSION)"; echo; \
	    [ -f $$TMPL ] && { cat $$TMPL; echo; }; \
	    echo "## Changes"; \
	    if [ -n "$$PREV" ]; then git log --oneline $$PREV..HEAD; else git log --oneline; fi; } > $$TMP; \
	  $${EDITOR:-vi} $$TMP; \
	  git tag -a "$(VERSION)" -F $$TMP; \
	  rm $$TMP
	@echo "Tag created locally. Run: git push origin $(VERSION)"
```

Two-phase intentional: `make release` creates tag locally only. Push is manual (`git push origin $VERSION`) so a fat-fingered version doesn't immediately fan out to remote + downstream.

### `.github/RELEASE_TEMPLATE.md`

```markdown
<!-- Edit before tagging. Auto-populated change list appended below. -->

## Highlights

-

## Breaking changes

None.

## Upgrade notes

-
```

### GoReleaser changelog (edgesync, bones)

```yaml
changelog:
  use: github
  sort: asc
  groups:
    - title: 'Features'
      regexp: '^.*?feat(\(.+\))??!?:.+$'
      order: 0
    - title: 'Bug fixes'
      regexp: '^.*?fix(\(.+\))??!?:.+$'
      order: 1
    - title: 'Other'
      order: 999
  filters:
    exclude:
      - '^docs:'
      - '^test:'
      - '^chore:'
      - '^ci:'
```

libfossil uses `gh release create --generate-notes` instead, which produces equivalent grouping from PR labels + commit prefixes.

## Rollout — 6 PRs, sequential

| # | Repo | Title | Adds | Risk |
|---|------|-------|------|------|
| 1 | bones | `chore: add make ci, ci-fast, release helper, pre-push hook` | Makefile targets, `scripts/install-hooks.sh`, `.github/RELEASE_TEMPLATE.md` | Low |
| 2 | edgesync | same as #1 | same | Low |
| 3 | libfossil | same as #1 (no goreleaser config exists yet to touch) | same | Low |
| 4 | bones | `feat(ci): add bump-upstream workflow + goreleaser changelog` | `bump-upstream.yml`, goreleaser changelog | Medium |
| 5 | edgesync | `feat(ci): add bump-upstream workflow, dispatch on tag, goreleaser changelog` | `bump-upstream.yml` + dispatch step in `release.yml` + goreleaser changelog | Medium-high |
| 6 | libfossil | `feat(ci): add release workflow with downstream dispatch` | new `release.yml` | Medium |

Order rationale:

- PRs 1-3 (local dev hygiene) are pure adds with no remote behavior change. Land them first to have `make ci` and pre-push hooks across all repos before touching dispatch wiring.
- PR 4 lets bones be tested as a receiver in isolation via manual `gh api` fire.
- PR 5 brings up both sides of the dispatch in one repo; tests both directions in one cycle (manual fire to receiver, real tag to test sender).
- PR 6 closes the chain. After this PR merges and a real libfossil release fires, the cascade is alive.

## PAT setup

One-time, before PR #4 merges:

1. **`DISPATCH_PAT` for libfossil** — fine-grained PAT, repo access: `danmestas/EdgeSync` only, permissions: `Contents: Read and write`. Stored in libfossil repo secrets.
2. **`DISPATCH_PAT` for edgesync** — fine-grained PAT, repo access: `danmestas/bones` only, same perms. Stored in `danmestas/EdgeSync` repo secrets.

1-year expiry; calendar reminder for rotation. Existing `HOMEBREW_PAT_TOKEN` is too broad to reuse.

## Testing strategy

### Per-PR local validation

PRs 1-3 (Makefile/hooks):
- `make ci` exits 0 on a clean main.
- `make ci-fast` runs in <60s.
- `make setup-hooks` installs `pre-push`; verify with `cat .git/hooks/pre-push` and `git push --dry-run`.
- `make release VERSION=v9.9.9-test` validates and tags locally; `git tag -d v9.9.9-test` to clean up.

### Receiver workflow validation (PR #4, then again in #5)

Manual fire after PR merge:

```bash
gh api -X POST repos/danmestas/bones/dispatches \
  -f event_type=upstream-bump \
  -f client_payload[repo]=EdgeSync \
  -f client_payload[module_path]=github.com/danmestas/EdgeSync \
  -f client_payload[version]=v0.0.7 \
  -f client_payload[release_url]=https://github.com/danmestas/EdgeSync/releases/tag/v0.0.7 \
  -f client_payload[tag_sha]=$(gh api repos/danmestas/EdgeSync/git/refs/tags/v0.0.7 -q .object.sha) \
  -f client_payload[actor]=danmestas
```

**Expected outcomes**:

| Scenario | Expected |
|----------|----------|
| First fire with already-pinned version | `noop=true`, no PR created |
| First fire with new version | PR opens on `chore/bump-EdgeSync-vX.Y.Z` |
| Re-fire same version | "PR already open" notice, no second PR |
| Bad version (`vbogus`) | Workflow fails with `::error::bad version` |
| Wrong repo (`repo=foobar`) | Workflow logs notice and exits 0 silently |
| `workflow_dispatch` with same inputs | Same behavior as `repository_dispatch` |

### End-to-end cascade validation (PR #5, PR #6)

After PR #5:
1. Tag edgesync with `v0.0.8-rc1`.
2. Watch `release.yml` fire goreleaser + dispatch step.
3. Watch bones receive dispatch and open PR `chore/bump-EdgeSync-v0.0.8-rc1`.
4. Close PR (don't merge); delete `v0.0.8-rc1` tag from origin and the prerelease GitHub release.

After PR #6:
1. Tag libfossil with `v0.4.6-rc1`.
2. Watch libfossil `release.yml` fire `gh release create` + dispatch.
3. Watch edgesync open PR.
4. Don't merge edgesync PR; close it; delete the rc tag and prerelease release.

`-rc1` suffix matters: changelog/release configs treat non-numeric suffixes as prereleases, so rc tags don't pollute the main release stream.

## Decisions log

| Decision | Choice | Rationale |
|----------|--------|-----------|
| Scope | Bucket C: local dev hygiene + dispatch automation + release ergonomics | Closes the loop on consistency; deferred CI backlog handled separately |
| Dispatch mechanism | `repository_dispatch` (not Dependabot) | Tag-triggered, seconds latency, customizable PR body with upstream changelog link |
| Pre-push enforcement | Fast subset (vet+build+short tests, ~60s); full mirror behind `make ci` | Hard-block on 7-min hook would train `--no-verify` bypass; fast hook catches cheap mistakes |
| Dispatch graph | Linear (libfossil→edgesync→bones); libfossil does NOT dispatch to bones directly | Avoids fan-in to bones (the original "two PRs" pain); accepts that bones needs an edgesync release to pick up libfossil-only updates |
| Rollout | 6 PRs (3 hygiene + 3 dispatch) | Per CLAUDE.md, every PR is reviewed manually — smaller diffs are easier to review |
| libfossil release pipeline | `gh release create --generate-notes` (not goreleaser) | Library, no binaries; goreleaser would be 50 lines of YAGNI |
| Spec location | `libfossil/docs/superpowers/specs/` | Matches prior `2026-04-25-docs-site-cf-deploy-design.md`; libfossil is the source of the chain |
| Bot identity | `github-actions[bot]` (uid 41898282) | GitHub stock identity; complies with "never attribute Claude/AI" rule |

## Open questions / future work

- **Cross-repo CI**: should an edgesync PR run libfossil's tests against its current pin? Possible follow-up; out of scope here.
- **Reusable workflow**: `bump-upstream.yml` is ~95% identical between edgesync and bones. Could move to a fourth `.github` repo as a reusable workflow once the pattern stabilizes. Premature for a 3-repo cascade.
- **Auto-merge tier**: if cascade reliability is high after a few real cascades, consider an `automerge` label that lets bot-authored chore-bump PRs merge on green CI. Requires branch protection + status checks. Not in this design.
- **Deferred CI improvements**: paths-ignore for docs PRs, `-race` in libfossil, SDK-drift check in libfossil, DST seed sweep. Tracked in `project_pending_ci_improvements.md`. Could fold into PRs 1-3 but bloats their scope; recommend keeping separate.
- **Diamond dep edge case**: linear-only means bones lags libfossil if edgesync doesn't release. If a libfossil-only fix becomes urgent for bones, manual override is `gh workflow run bump-upstream.yml -f repo=libfossil ...` after temporarily relaxing the `EXPECTED_REPO` check, or a one-off manual go.mod bump. Acceptable today; revisit if it happens more than annually.