[MarkusNeusinger]

Summary

• impl-generate.yml and bulk-generate.yml get a source_ref input (default empty). With it set, impl-generate takes plots/{spec}/implementations/{language}/{library}{ext} from that branch, tag or SHA of origin instead of running Claude. A new step, "Adopt implementation from source ref", renders it in both themes with the generator's Step 3 commands (Python MPLBACKEND=Agg python -P, the JS render harness), checks both PNGs with core.canvas, and commits and pushes that one file. Every later step runs unchanged: metadata, adaptability smoke run, image processing, PR, staging upload, preview, review dispatch. The purpose is to let a gemini/* branch that migrates the 323 MUI X implementations to MUI X 9 reach main only through render, Opus review, repair and impl-merge.
• An adopted run is always a forced regeneration: the REGEN_GATE expression (the same in all three steps) treats source_ref != '' like regen_gate=false, so the PR is labelled regen:forced. Both Claude steps and the prompts overlay are skipped with if:. The routed model only runs repair. The PR body adds a **Generation:** adopted from <ref> (<sha>) line and keeps the **Model:** line, because the rescue paths of impl-review and impl-repair parse exactly that line. The metadata records generated_by: adopted:<ref>.
• A source failure (the file or the ref is missing, the render fails, or the pair is R or Julia) writes cause before the step fails. The failure handler then posts one issue comment with its own marker (impl-adopt-fail). It does not auto-retry, counts no attempt and never sets impl:<lib>:failed, because a retry would read the same file. Any later failure takes the ordinary path, and dispatch_retry now forwards source_ref.

Plan

N/A. The design follows the lead's contract. Decisions are listed in the test plan notes below.

Test plan

• New tests/unit/workflows/test_impl_generate_adopt.py (77 cases). It runs the real adopt step against a throwaway origin under bash -e and bash -eo pipefail, covering: Python and JS renders, branch/tag/SHA refs, a file identical to main (empty adopt commit), canvas drift (warns, still pushes), a missing file or ref, a failing or silent or non-PNG render, R/Julia refused, numpy installed into the helper venv, and staged unrelated changes kept out of the commit. It also runs "Extract inputs" (normalization of origin/, validation, notices), the metadata writer (generated_by), the failure handler (source failure vs later failure, retry forwarding), and bulk-generate's dispatch loop. Wiring guards cover if: conditions, step order and the shared REGEN_GATE expression.
• test_model_routing.py: the adopted PR body round-trips through impl-review's "Extract PR info" and impl-repair's "Resolve model" rescues (model opus survives).
• uv run ruff check ., ruff format --check ., mypy api core agents, uv run pytest tests/unit -n 4 (13240 passed), and tools.changelog check --base origin/main. actionlint is not installed locally, so it was not run.
• Real run from this branch: impl-generate run 39976443391, gauge-activity-rings / muix / source_ref=main (a spec whose muix.tsx renders identically on MUI X 9). The adopt step succeeded, the Claude steps and the overlay were skipped, and canvas_gate passed at 2400x2400 in both themes. PR feat(muix): implement gauge-activity-rings #12273 opened with regen:forced, the adopted-from line, generated_by: adopted:main and library_version: 9.16.0, and its review dispatched. impl-review, impl-repair and impl-merge are unchanged, so their Claude steps pass claude-code-action's workflow validation on the branch ref.

[Copilot]

🟡 Changes recommended

Unresolved workflow defects include a critical dependency-install failure path and multiple moderate adoption validation, fetch, failure-notification, and regeneration-gate issues.

3 open issues

• Make dependency installation and NumPy validation fatal · New
• Base REGEN_GATE on normalized source_ref · New
• Distinguish fetch failures from missing source refs · New

What changed in this PR

This pull request adds an adoption mode that lets implementation files from a branch, tag, or commit SHA pass through rendering, validation, review, repair, and merge without initial Claude generation.

Changes:

• Adds source_ref handling to the single and bulk generation workflows.
• Adds source rendering, canvas validation, metadata, PR-body, retry, and failure-reporting support.
• Adds workflow tests and documents adoption behavior, routing, and operational usage.

File	Summary	Review notes
tests/​unit/​workflows/​test_regen_gate_workflows.py	Updates regeneration and failure-handler coverage.	No final comments.
tests/​unit/​workflows/​test_model_routing.py	Verifies adopted PR model metadata survives review and repair rescue paths.	No final comments.
tests/​unit/​workflows/​test_impl_generate_adopt.py	Tests adoption rendering, validation, metadata, commits, failures, retries, and bulk wiring.	No final comments.
docs/​workflows/​overview.md	Documents adoption mode, forced regeneration, and failure behavior.	Nit: the documented 323-item migration exceeds the bulk workflow’s 360-minute timeout at its default pacing.
changelog.d/​impl-generate-adopt.md	Adds the changelog entry for the new capability.	No final comments.
agentic/​docs/​project-guide.md	Documents adoption behavior in the pipeline and regeneration architecture.	No final comments.
.github/​workflows/​impl-generate.yml	Implements source adoption, rendering, canvas checks, metadata, PR creation, retry, and failure handling.	Moderate: raw source_ref=origin/ can force regeneration despite normalization. Critical: failed dependency installation can be treated as tolerated canvas drift. Moderate: adoption failures without an issue receive neither notification nor cleanup. Moderate: fetch failures are treated as missing refs and suppress retries. Moderate: invalid refs follow the ordinary retry path and consume attempts.
.github/​workflows/​bulk-generate.yml	Passes source_ref through bulk dispatches and reports adoption mode.	No final comments.
.claude/​skills/​babysit-pipeline/​SKILL.md	Documents monitoring and retry behavior for adopted runs.	No final comments.

_{Review effort: Lite}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!