Document gh pr create workaround for restricted fetch refspecs

[dmccoystephenson]

Summary

• Adds an Edge-case entry to create-dev-loop.md documenting a workaround for a gh pr create failure seen in gardener-managed (and similarly restricted) clones: when origin's fetch refspec is limited to the default branch only, git fetch never populates a remote-tracking ref for feature branches, so gh pr create and git branch --set-upstream-to both fail their auto-detection even after a successful git push -u. The fix: pass --head <branch> explicitly.
• This propagates the workaround to every generated <slug>-dev-loop skill going forward, since the failure mode is environmental (restricted clone), not repo-specific.

Closes #88

Deferred (this cycle)

Three other open issues were triaged and intentionally left untouched:

• [self-audit] create-dev-loop-dev-loop's configured self-audit tracker repo doesn't exist #87, README and CLAUDE.md link gardener as an open-source companion, but the repo is private #79 — both require editing CLAUDE.md, which this loop treats as agent-loaded config requiring explicit human authorization to edit.
• [self-audit] create-dev-loop-dev-loop's configured self-audit tracker repo doesn't exist #87 additionally concerns create-dev-loop-dev-loop's own configuration (self-audit tracker repo doesn't exist) — Phase 9 forbids implementing changes to the skill itself; a human must review and approve skill edits.
• CLAUDE.md's documentation sources-of-truth table omits .github/ISSUE_TEMPLATE/*.md #81 also requires a CLAUDE.md edit for the same reason.

Research grounding

This is a mechanical process/edge-case fix (a gh CLI behavior workaround), not a change to a phase definition or the core template logic — no RESEARCH.md finding applies.

Test plan

• python3 scripts/check_docs.py passes (placeholder/table/link consistency).
• Confirmed no new {{placeholder}} was introduced (plain prose addition), so no Step 4 substitution-table row is needed.
• Confirmed no fenced code block was added (no backslash-escaping needed).
• Confirmed the new bullet's structure/tone matches sibling Edge-case entries (bold lead-in phrase + colon + description).
• UNVERIFIED — full /create-dev-loop fixture regeneration against a reference repo (e.g. dpm-dev-loop) was not run: the harness sandbox for this session denies filesystem access outside this repo's working directory (~/local-skills is blocked), so the external validation anchor cannot run this cycle. Since this PR modifies template surface (create-dev-loop.md), full validation is required per this loop's own gating rules — flagging UNVERIFIED and not auto-merging. create-dev-loop.md is also already on the do-not-auto-merge path list (any template change deserves careful human review), so this is consistent either way. A human (or a session with unrestricted filesystem access) should run the Step 1-5 regeneration against a fixture repo and confirm the new Edge-case text renders correctly before merging.

Self-audit followup

Filed #90 during this cycle's Phase 9 self-audit: Phase 4's anchor-cannot-run guidance doesn't name session-sandbox path restriction (the case hit in this very PR's validation) as a trigger alongside "tool/interpreter absent or broken".

🤖 Generated with Claude Code

---

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

[dmccoystephenson]

Self-review rubric:

• Scope: PASS — single Edge-case bullet added, nothing else touched; directly resolves [self-audit] gardener-managed clones' restricted fetch refspec breaks gh pr create after git push -u #88's named surface area.
• Tests-new: N/A — no new step/placeholder added, so no new manual validation step is required. check_docs.py re-run to confirm no placeholder/table drift was introduced.
• Tests-fix (empirical): PASS-partial — the underlying gh pr create failure was already reproduced and captured verbatim in [self-audit] gardener-managed clones' restricted fetch refspec breaks gh pr create after git push -u #88's report (from a real gardener-dispatched cycle), so the defect this text documents is independently confirmed, not asserted from reasoning. Stash-and-revalidate isn't meaningful here since this is prose guidance, not template logic with an executable failure mode to toggle.
• Sibling structure: PASS — new bullet follows the exact **<condition>:** <guidance> format used by every neighboring Edge-case entry.
• Sibling renames: N/A — no identifiers renamed.
• Docs: PASS — python3 scripts/check_docs.py passes; no {{placeholder}} introduced so no Step 4 table row needed; no fenced code block added so no escaping needed.
• Issue resolution: PASS — [self-audit] gardener-managed clones' restricted fetch refspec breaks gh pr create after git push -u #88's suggested action ("Add this as a named workaround in create-dev-loop.md's Phase 4 ... or Edge cases") is implemented verbatim in Edge cases, including the --head <branch> workaround and the diagnostic command.
• Manual validation: UNVERIFIED — this session's harness sandbox denies filesystem access outside this repo's working directory (~/local-skills is blocked), so the full /create-dev-loop fixture regeneration this loop's Phase 4 anchor calls for cannot run this cycle. Since this PR touches template surface (create-dev-loop.md), full validation is required per this loop's own gating rules but is unavailable here — flagging UNVERIFIED rather than claiming green, and not auto-merging as a result.
• Placeholder-consistency: PASS — no placeholders added or removed.
• Escaping: N/A — no fenced code block added.
• Research-grounding: N/A — this is a mechanical gh CLI edge-case fix, not a change to a phase definition or core template logic; no RESEARCH.md finding applies (stated in the PR body).

Summary: small, low-risk, well-grounded addition (the failure it documents was directly observed, not hypothesized). Leaving open for human merge rather than auto-merging, both because the fixture anchor is UNVERIFIED this cycle and because create-dev-loop.md changes are on this repo's do-not-auto-merge path list regardless.