# Synapseml Pr Loop > Make one or more SynapseML issues or pull requests evidence-based merge-ready. Use for "5/5 confidence", "200% ready", stale/outdated PR remediation, rebase-and-test requests, resolving all review comments, or proving a feature ships without correctness, compatibility, performance, or Spark regressions. - Skill: `microsoft/synapseml-pr-loop` (Agent Skill, multi-file: 5 files) - Install (CLI): `npx skillmds@latest add microsoft/synapseml-pr-loop` - Raw SKILL.md: https://api.skillmd.com/api/skills/microsoft/synapseml-pr-loop/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Microsoft (https://skillmd.com/u/microsoft) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/microsoft/synapseml-pr-loop --- # SynapseML PR loop Treat "5/5" or "200%" as an evidence standard, never a literal guarantee. The exit condition is: the requested value is proven through the public API, the current target is integrated, review is exhausted, and every required check is complete and green. ## Workflow ### 1. Establish scope and isolation - Load the [branch context skill](../synapseml-branches/SKILL.md) using the PR base branch. Recheck it before validation and immediately before final push. - Read the issue, PR body, linked work items, commit history, changed files, and every review thread/body, including resolved, outdated, minimized, and suppressed comments. Verify prior resolutions rather than trusting status. - Inspect formal review decisions, requested-change votes, ownership gates, and coverage thresholds; resolved threads do not clear those blockers. - Check recently merged/closed related PRs and issues. Identify follow-up PRs needing rebase/remediation, superseded work to close, and remaining issue action items; do not assume closure completed the feature lifecycle. - Give each PR a dedicated worktree and branch. Parallelize independent PRs, but identify overlapping files and required merge order first. - Run [scripts/Get-PrReadiness.ps1](scripts/Get-PrReadiness.ps1) with `-PullRequest ` and retain its JSON locally as the initial snapshot. It can contain review text; redact it before public sharing. ### 2. Integrate the current target - Fetch the PR's target branch and rebase an ordinary PR before validation. - Use `--force-with-lease`, never an unguarded force push. - Merge, rather than rebase, shared `spark` port branches. - Record target SHA, head SHA, merge base, ahead/behind counts, and conflicts. - Compare the intended patch before and after rebase/conflict resolution. - Fetch again immediately before the final push. If the target advanced, integrate it and rerun affected validation. ### 3. Define the value and regression contract - Keep the PR title and description aligned with the current scope. Lead with a short human-readable change/value summary; put detailed design and validation evidence afterward. Refresh both after material changes. - State the user-visible bug or feature, supported/unsupported cases, default behavior, compatibility contract, and measurable acceptance criteria. - Trace the real public path: Scala stage, generated/hand-written Python, schema, serialization, persistence, service/native boundary, and packaging. - Confirm the published package actually contains the capability; local jars, custom natives, or provider discovery do not prove that users receive it. - Establish a baseline when failures, performance, or external systems are involved. A passing new test is insufficient if the old behavior was never shown to fail. ### 4. Review and implement - Apply the [code-review skill](../code-review/SKILL.md). - Resolve root causes, not only the reported line. Recheck sibling APIs and language surfaces that share the same serializer, schema, parameter, or native/service path. - Preserve public JVM and serialized compatibility unless explicitly approved. - Update user-facing documentation/examples for changed public behavior. Edit Scala sources rather than generated files under `target/`. - Follow the Spark and performance gates in [references/spark-performance.md](references/spark-performance.md). - Reply in the existing thread with the fix and evidence, then resolve it. - Re-audit after every push. Automated review is asynchronous and re-runs per commit, so auditing immediately after pushing reads the *previous* review and reports a false all-clear. Wait until the newest automated review's commit equals the pushed head, then audit; poll rather than checking once. - Suppressed comments are not review threads. They appear only inside a collapsed section of the review body, so a `reviewThreads` query returns zero while they exist, and they have no thread to reply to or resolve. Read every automated review body for the current head, and address them in the follow-up commit message or a PR comment. Treat them as ordinary findings: they are suppressed for confidence, not for correctness. ### 5. Add proof-oriented tests - Add a regression that fails before the fix and passes after it. - Cover positive, negative, null/empty, boundary, schema, copy, save/load, and Python/codegen behavior as applicable. - Exercise the public transformer/estimator or request path end to end; helper tests alone do not prove the feature ships. - Use real hardware, native libraries, clusters, network families, or services when the claim depends on them. Do not infer capability from configuration or provider discovery alone. - Before external service tests, audit resource creation/deletion and use only authorized test resources. ### 6. Validate locally and across branches - Use the [local setup skill](../synapseml-local-setup/SKILL.md) and its JDK wrapper. - Run the smallest targeted suites, compile, test compile, Scala style, pinned Black, codegen, generated-wrapper checks, and relevant Python tests. - Run release compatibility for every port branch affected by the change. - Benchmark representative scale before/after when a hot path, network path, accelerator, allocation pattern, or algorithmic complexity changes. ### 7. Run and triage full CI - Push the exact validated head, comment `/azp run`, then confirm a build actually queued -- a comment is not evidence that CI ran, so cite the build ID. A trigger-driven build records `reason=pullRequest`; one you queued yourself records `reason=manual`, which is the quickest way to tell whether the trigger really fired or you merely re-ran it by hand. - Do this after **every** push, not once per pull request. The build does not re-queue itself when the head moves, so the previous run's result belongs to code that no longer exists. The GitHub Actions checks do re-run on each push and go green within a couple of minutes, which makes a head with no Azure Pipelines build on it look fully checked; an absent check is neither failed nor pending, so nothing reports it. Verify the build against the head SHA by name, or run `Get-PrReadiness.ps1 -RunPipeline` to post the comment automatically when it is missing. - If no build appears, check the pipeline definition's own pull-request trigger rather than assuming a transient failure. That trigger can be defined in the pipeline UI, in which case it overrides the `pr:` block in `pipeline.yaml` entirely and silently ignores targets the YAML lists. Read its branch filters through the definitions API. Until the filter is corrected, queue explicitly against `refs/pull//merge` -- never `refs/heads/`, which validates the branch instead of the merge result. - Inspect every failed, canceled, skipped, and pending job. Use [references/ci-triage.md](references/ci-triage.md) to separate product defects, test defects, baseline failures, and infrastructure failures. - Fix product/test defects and rerun. Infrastructure classification requires logs proving tests did not exercise the change; "looks flaky" is not evidence. - If path filters or a CI-only diff bypass the behavior being repaired, validate it with a representative product change or controlled integration PR. - Do not declare readiness while any required check is pending. ### 8. Final readiness loop Run `Get-PrReadiness.ps1 -PullRequest -WaitForReview -RunPipeline` after the final push and confirm every gate in [references/readiness-gates.md](references/readiness-gates.md). Those two switches cover the asynchronous gaps that a bare snapshot reports as clean: the automated review has not arrived yet, and the Azure Pipelines build has not been asked to start. Both leave the same signature -- nothing failed, nothing pending, nothing there. For multiple PRs, after each merge: 1. fetch the new target; 2. rebase overlapping downstream PRs; 3. rerun targeted, compatibility, and full CI; 4. re-audit review threads and suppressed comments. After any merge or closure, reconcile linked work: update or close fulfilled issues, close superseded PRs with an explanation, and rebase/remediate still valuable follow-ups. Preserve separate unresolved scope rather than closing it for convenience. Report the exact remaining blocker. "Only human approval remains" is valid only when all engineering gates are complete.