Skip to content

Check the default branch's current CI state at Phase 1 - #178

Merged
dmccoystephenson merged 3 commits into
mainfrom
feature/check-default-branch-ci
Nov 15, 2026
Merged

dmccoystephenson merged 3 commits into
mainfrom
feature/check-default-branch-ci

Conversation

@dmccoystephenson

@dmccoystephenson dmccoystephenson commented Nov 15, 2026 •

Copy link
Copy Markdown
Member

Summary

  • A paragraph is added to the template's Phase 1, right after the structurally-red check. It tells a generated loop to read the latest run of each required workflow on the default branch. If that run failed, the breakage becomes the cycle's work ahead of the backlog, because every PR branched from a red base inherits the red check and the Phase 4 anchor stops saying anything about the change.
  • The same paragraph has the loop run the Phase 3 verify commands once on an unmodified default branch before any edit, even when the branch is green. A failure first seen after editing can't be told apart from one the edit caused.
  • The baseline points at "the Phase 3 verify commands" rather than repeating {{COMPILE_CMD}}. That placeholder can resolve to a shell comment in docs-only repos, and the reference reads correctly either way. No new placeholder is introduced and no phase is renumbered.
  • A RESEARCH.md §1 Implementations entry is added. Its PR number and ship date are filled in after merge.

Closes #144

Retrofit note: this adds a rule to the template, so every existing <slug>-dev-loop skill generated before this change lacks it. That includes create-dev-loop-dev-loop (template-version 44756ec) and medieval-factions-dev-loop, where the issue was first observed (tracked there as #51). Each of these probably needs a retrofit pass. Only the operator can see the full list of installed skills.

Skipped issues: all other open issues were deferred to keep this PR to one change. #173 and #106 need CLAUDE.md edits, which require separate authorization. #109 and #114 are human-gated verification questions. The remaining template-rule issues are independent template changes left for later cycles.

Research grounding

RESEARCH.md §1 (self-critique needs an external signal). PR #121 already treats an anchor whose required job has never passed as carrying no signal. This change covers the transient version: a job that passes routinely but is red right now gives just as little signal for the PR under review. The issue reports this from a generated instance where main had failed its Build check for five days without any cycle noticing.

Doc sync check

  • README.md's "What it does" Step list still matches create-dev-loop.md's Steps 1:1 (no Step changed)
  • Every {{placeholder}} added or changed has a corresponding Step 4 substitution-table row (only the existing {{DEFAULT_BRANCH}} is used)
  • RESEARCH.md updated (an Implementations entry under §1)

Test plan

  • python3 scripts/check_docs.py printed "Doc consistency check passed."
  • python3 -m unittest discover -s tests ran 42 tests, OK
  • UNVERIFIED: /create-dev-loop was not run against a real repo. This headless dispatch isn't authorized to create a GitHub repo (Step 6) or write to ~/local-skills/ / ~/.claude/commands/, so none of the five CLAUDE.md "Testing changes" items were exercised. The change is one added prose paragraph with no new placeholder, so the main risk (an unresolved placeholder) is covered by check_docs.py. The paragraph's behavior in a generated skill has not been checked.

This PR description was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

🤖 Generated with Claude Code


drafted by Claude on behalf of Daniel Stephenson

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@dmccoystephenson

Copy link
Copy Markdown
Member Author

Self-review rubric (performed inline, since the dispatch has no reviewer subagent):

  • Scope: PASS. Only create-dev-loop.md (one Phase 1 paragraph) and RESEARCH.md (one §1 Implementations entry) changed, and both are needed for Phase 1 triage never checks whether the default branch's own CI is green #144.
  • Tests-new: PASS (n/a). No function was added. scripts/check_docs.py is untouched.
  • Tests-fix: PASS (n/a). This adds a template rule rather than fixing a code bug, so no regression test can reach it.
  • Sibling structure: PASS. The new paragraph follows the bold-lead-sentence form of the Phase 1 paragraphs around it.
  • Sibling renames: PASS (n/a). Nothing was renamed.
  • Docs: PASS. README's Step list is unaffected because no Step changed. SECURITY.md is unaffected because Step 2's read list didn't change. The CONTRIBUTING.md and PR/issue templates don't restate Phase 1's contents.
  • Issue resolution: PASS. Both halves of Phase 1 triage never checks whether the default branch's own CI is green #144's suggested text are covered: the current-red check and the pre-edit baseline.
  • CI: PASS. doc-consistency passed on the head commit. Locally, check_docs.py passed and the unit tests ran 42, OK.
  • Placeholder parity: PASS. Only the existing {{DEFAULT_BRANCH}} is used, and check_docs.py passed.
  • README Step parity: PASS (n/a).
  • Fence escaping: PASS. No fence was added.
  • Phase numbering: PASS. Nothing was renumbered. The new paragraph's "Phase 3" and "Phase 4" references point at the verify commands and the external anchor as they exist now.
  • Version comments: PASS (n/a).
  • Research grounding: PASS. The body cites §1, and an Implementations entry was added with PR #TBD / shipped pending. Those placeholders must be replaced with Check the default branch's current CI state at Phase 1 #178 and the mergedAt date once merged.
  • No back-ported specifics: PASS. The Build check and Medieval-Factions appear only in RESEARCH.md as evidence, not in the template.
  • gardener interface: PASS (n/a). Steps 3, 5 and 6 didn't change.
  • Retrofit note: FAIL at first, now fixed. The first PR body didn't say which existing skills need a retrofit. A retrofit note has since been added to the body.
  • Behavioral validation: FAIL. No /create-dev-loop run was done, because the dispatch isn't authorized for Step 6 / ~/local-skills/ side effects, so the change is marked UNVERIFIED.

Judgment call worth flagging: the new text calls a red default branch "this cycle's work" that outranks the backlog. The run may not be able to fix every such breakage, for example a flaky external dependency. In that case the existing structurally-red / hand-off paths would apply, but the paragraph doesn't say so.

Summary: CI passes and the change is mechanically consistent. The template behavior is UNVERIFIED, so under the skill's own Behavioral-validation gate this PR is handed off for human merge rather than merged automatically.

This review comment was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).


drafted by Claude on behalf of Daniel Stephenson

dmccoystephenson and others added 2 commits November 14, 2026 23:36
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@dmccoystephenson
dmccoystephenson merged commit a9ac0bf into main Nov 15, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Phase 1 triage never checks whether the default branch's own CI is green

1 participant