You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[#850] repo-template baseline re-seed pins BARE @dev-lead/stable — reverts the v<M>-stable convergence (flapping) #886
The repo-template baseline re-seed (chore: seed org baseline file scaffold, e.g. c917e6d7 / PR repo-template#86, and the earlier 707c9a49) rewrites repo-template's .github/workflows/dev-lead.yml back to the bare@dev-lead/stable pin (+ agent_ref: dev-lead/stable), undoing the standards-deploy convergence to @dev-lead/v1-stable. So repo-template flaps: standards-deploy re-pins it → re-seed reverts it. Since repo-template is the SEED for new repos, new repos then start on the bare tag (drift).
Fix: the scaffold source the re-seed copies from must pin @dev-lead/v1-stable (matching the standards template, incl the S7635/S7637 markers on secrets:inherit/uses lines). Verify the re-seed uses the same standards/workflows/dev-lead.yml (post-#860/#876) rather than a stale embedded copy. Also spot-check the other stubs the re-seed writes (auto-rebase, dependabot-automerge, etc.) for the same bare-pin/missing-marker regression.
Surfaced by Epic #850 / #857 fleet convergence + verification. Blocks a clean compliance audit (repo-template shows as wrong-channel-pin).
Issue:#886 — [#850] repo-template baseline re-seed pins BARE @dev-lead/stable — reverts the v-stable convergence (flapping)
Root cause (verified)
repo-template@c917e6d7 (PR fix: encode compliance-fix learnings into standards and Claude prompt #86, "chore: seed org baseline file scaffold") rewrote the correct stubs to bare tier pins — the diff flips @dev-lead/v1-stable → @dev-lead/stable (and agent_ref), @auto-rebase/v2-stable → @auto-rebase/stable, @dependabot-automerge/v2-stable → @dependabot-automerge/stable, and drops the trailing newline. This is a stale embedded copy, not standards/workflows/.
standards/workflows/*.yml in this repo are already correct (v<M>-stable + S7637/S7635 markers) and guarded by test/scripts/standards-templates/.
PR fix: encode compliance-fix learnings into standards and Claude prompt #86 attributes generation to seed-repo-template.sh, but that script currently only knows --emit-baseline .gitignore — it has no path to emit the workflow stubs from the canonical standards/workflows/, so the re-seed fell back to an embedded bare-pinned copy. That missing path is the fix surface in this repo.
Scope
Teach seed-repo-template.sh to emit workflow-stub baseline files verbatim from standards/workflows/ (single source of truth), with a defensive guard that refuses to emit a bare <agent>/<tier> pin — so the re-seed can never again re-seed repo-template onto a wrong-channel tag.
Implementation Checklist
scripts/seed-repo-template.sh: add --emit-workflow <name.yml> that prints standards/workflows/<name.yml> verbatim; assert its first-party channel-ref pin is major-scoped v<M>-<tier> (never bare) and exit non-zero on regression or unknown name.
.github/workflows/gitignore-baseline-tests.yml: register the new bats file in paths: + the bats invocation so CI gates it.
Tests to Write
tests/test_seed_repo_template_workflows.bats:
--emit-workflow dev-lead.yml is byte-identical to standards/workflows/dev-lead.yml.
emitted dev-lead/auto-rebase/dependabot-automerge stubs pin @<agent>/v<M>-stable (never bare @<agent>/stable) and carry the S7637 (uses:) + S7635 (secrets: inherit) markers.
unknown workflow name → non-zero exit.
a template with a bare pin → non-zero exit (defensive guard; exercised via a temp STANDARDS_DIR fixture).
Test Command
bats tests/test_seed_repo_template_workflows.bats (plus full suite + .dev-lead/scripts/dev-lead-lint.sh + shellcheck in Verify)
The chore: seed org baseline file scaffold re-seed (repo-template c917e6d7, PR #86) rewrote the correct stubs to bare tier pins — the actual diff flips @dev-lead/v1-stable → @dev-lead/stable (+ agent_ref), @auto-rebase/v2-stable → @auto-rebase/stable, @dependabot-automerge/v2-stable → @dependabot-automerge/stable, and drops the trailing newline. That is a stale embedded copy, not standards/workflows/. PR #86 attributes generation to seed-repo-template.sh, but that script only knew --emit-baseline .gitignore — it had no path to emit workflow stubs from the canonical standards/workflows/, so the re-seed fell back to an embedded bare-pinned copy. That missing path is the fix surface in this repo.
Fix
Taught seed-repo-template.sh to emit workflow-stub baseline files verbatim from standards/workflows/ (single source of truth) via a new --emit-workflow <name.yml> mode, with a fail-loud guard that refuses to emit any stub whose first-party channel-ref pin is a bare <agent>/<tier> (missing the v<M>- major prefix the compliance audit requires). The re-seed now can never re-seed repo-template onto a wrong-channel tag, and the verbatim copy also restores the trailing newline #86 dropped.
standards/workflows/*.yml were already correct (v<M>-stable + S7637/S7635 markers, guarded by test/scripts/standards-templates/); no template change was needed — only the generator path to reach them.
tests/test_seed_repo_template_workflows.bats: 7 tests (TDD — red before impl, green after)
.github/workflows/gitignore-baseline-tests.yml: register the new bats file in paths: + the bats invocation
Test Results
### bats: gitignore-baseline gate (56/56) ###
ok 50 --emit-workflow dev-lead.yml is byte-identical to the standards template
ok 51 emitted dev-lead stub pins @dev-lead/v1-stable, never the bare tier tag
ok 52 every re-seeded stub emits a major-scoped v<M>-stable pin (no bare tier)
ok 53 every re-seeded stub keeps its S7637 (uses:) and S7635 (secrets: inherit) markers
ok 54 --emit-workflow rejects an unknown workflow name
ok 55 --emit-workflow refuses a template carrying a bare <agent>/<tier> pin
ok 56 usage is shown when --emit-workflow is given no name
### bats: standards-templates gate (7/7) — unchanged, still green ###
### shellcheck --severity=warning -x scripts/seed-repo-template.sh — OK ###
### .dev-lead/scripts/dev-lead-lint.sh — all checks passed ###
Spot-check: --emit-workflow was run against all 18 standards/workflows/*.yml — every one emits cleanly (no false-positive refusals; the 4 verbatim non-channel templates included).
Files Changed
scripts/seed-repo-template.sh: new --emit-workflow <name.yml> mode + bare-pin guard
tests/test_seed_repo_template_workflows.bats: new bats suite (7 tests)
.github/workflows/gitignore-baseline-tests.yml: gate the new tests in CI
Notes
Out of scope (intentionally not touched): repo-template itself (external repo; constraints forbid cross-repo PRs — the corrected generator + the existing standards-sync sweep converge it), and the broader standards/ci-standards.md prose drift (the doc predates the v<M>- major-scoping org-wide and is a separate cleanup, not the #886 scaffold regression).
The new emit path mirrors the existing --emit-baseline contract (both invoked by the re-seed process), so this is not dead code — it is the canonical, standards-sourced generator the re-seed must call for workflow stubs.
Activity
don-petry commentedon Jul 23, 2026
Dev-Lead Implementation Plan
Issue: #886 — [#850] repo-template baseline re-seed pins BARE @dev-lead/stable — reverts the v-stable convergence (flapping)
Root cause (verified)
c917e6d7(PR fix: encode compliance-fix learnings into standards and Claude prompt #86, "chore: seed org baseline file scaffold") rewrote the correct stubs to bare tier pins — the diff flips@dev-lead/v1-stable→@dev-lead/stable(andagent_ref),@auto-rebase/v2-stable→@auto-rebase/stable,@dependabot-automerge/v2-stable→@dependabot-automerge/stable, and drops the trailing newline. This is a stale embedded copy, notstandards/workflows/.standards/workflows/*.ymlin this repo are already correct (v<M>-stable+ S7637/S7635 markers) and guarded bytest/scripts/standards-templates/.seed-repo-template.sh, but that script currently only knows--emit-baseline .gitignore— it has no path to emit the workflow stubs from the canonicalstandards/workflows/, so the re-seed fell back to an embedded bare-pinned copy. That missing path is the fix surface in this repo.Scope
Teach
seed-repo-template.shto emit workflow-stub baseline files verbatim fromstandards/workflows/(single source of truth), with a defensive guard that refuses to emit a bare<agent>/<tier>pin — so the re-seed can never again re-seed repo-template onto a wrong-channel tag.Implementation Checklist
scripts/seed-repo-template.sh: add--emit-workflow <name.yml>that printsstandards/workflows/<name.yml>verbatim; assert its first-party channel-ref pin is major-scopedv<M>-<tier>(never bare) and exit non-zero on regression or unknown name..github/workflows/gitignore-baseline-tests.yml: register the new bats file inpaths:+ the bats invocation so CI gates it.Tests to Write
tests/test_seed_repo_template_workflows.bats:--emit-workflow dev-lead.ymlis byte-identical tostandards/workflows/dev-lead.yml.@<agent>/v<M>-stable(never bare@<agent>/stable) and carry the S7637 (uses:) + S7635 (secrets: inherit) markers.Test Command
bats tests/test_seed_repo_template_workflows.bats(plus full suite +.dev-lead/scripts/dev-lead-lint.sh+ shellcheck in Verify)don-petry commentedon Jul 23, 2026
Dev-Lead: Implementation Complete
Root cause (verified against live repos)
The
chore: seed org baseline file scaffoldre-seed (repo-templatec917e6d7, PR #86) rewrote the correct stubs to bare tier pins — the actual diff flips@dev-lead/v1-stable→@dev-lead/stable(+agent_ref),@auto-rebase/v2-stable→@auto-rebase/stable,@dependabot-automerge/v2-stable→@dependabot-automerge/stable, and drops the trailing newline. That is a stale embedded copy, notstandards/workflows/. PR #86 attributes generation toseed-repo-template.sh, but that script only knew--emit-baseline .gitignore— it had no path to emit workflow stubs from the canonicalstandards/workflows/, so the re-seed fell back to an embedded bare-pinned copy. That missing path is the fix surface in this repo.Fix
Taught
seed-repo-template.shto emit workflow-stub baseline files verbatim fromstandards/workflows/(single source of truth) via a new--emit-workflow <name.yml>mode, with a fail-loud guard that refuses to emit any stub whose first-party channel-ref pin is a bare<agent>/<tier>(missing thev<M>-major prefix the compliance audit requires). The re-seed now can never re-seed repo-template onto a wrong-channel tag, and the verbatim copy also restores the trailing newline#86dropped.standards/workflows/*.ymlwere already correct (v<M>-stable+ S7637/S7635 markers, guarded bytest/scripts/standards-templates/); no template change was needed — only the generator path to reach them.Plan Execution
scripts/seed-repo-template.sh: add--emit-workflow(verbatim fromstandards/workflows/) +_assert_vform_pinsbare-pin guard + path-traversal/unknown-name rejectiontests/test_seed_repo_template_workflows.bats: 7 tests (TDD — red before impl, green after).github/workflows/gitignore-baseline-tests.yml: register the new bats file inpaths:+ the bats invocationTest Results
Spot-check:
--emit-workflowwas run against all 18standards/workflows/*.yml— every one emits cleanly (no false-positive refusals; the 4 verbatim non-channel templates included).Files Changed
scripts/seed-repo-template.sh: new--emit-workflow <name.yml>mode + bare-pin guardtests/test_seed_repo_template_workflows.bats: new bats suite (7 tests).github/workflows/gitignore-baseline-tests.yml: gate the new tests in CINotes
standards-syncsweep converge it), and the broaderstandards/ci-standards.mdprose drift (the doc predates thev<M>-major-scoping org-wide and is a separate cleanup, not the#886scaffold regression).--emit-baselinecontract (both invoked by the re-seed process), so this is not dead code — it is the canonical, standards-sourced generator the re-seed must call for workflow stubs.feat: implement issue #886 — [#850] repo-template baseline re-seed pi…
feat(standards): per-repo initiative-driver caller stub (#884) (#523)
feat(standards): per-repo initiative-driver caller stub (#884) (#523)
feat(standards): per-repo initiative-driver caller stub (#884) (#523)