SWE PR Risk Review

What This Skill Does

Use this skill to review an open or draft pull request before merge and decide where the engineering risk is concentrated.

The expected outcome is a concise risk report that tells the user:

• what the PR changes
• what makes it risky or low-risk
• what validation is missing or weak
• what rollout, rollback, or migration gaps still need attention
• what the smallest safe next action is

This skill is deliberately narrower than post-merge monitoring and bug hunting. It is about merge readiness, not production impact and not retrospective defect mining.

When To Use

Use this skill when the user wants to:

• review a PR before merge
• decide whether a change is safe to merge
• identify missing tests, checks, or rollout planning on an open PR
• assess whether migrations, flags, or interface changes need extra caution
• get agent-safe follow-up actions without doing the merge

Do Not Use

Do not use this skill for:

• post-merge production monitoring
• scanning recent commits for regressions after the fact
• a broad code style or architecture review with no PR scope
• general project planning or roadmap feedback
• security-specific review when the user explicitly asked for a security audit

Inputs To Confirm

Confirm or infer:

• the PR number, branch, or diff scope
• the repository or service in scope
• whether CI or local validation data is available
• whether the user wants report-only output or a concrete follow-up plan
• any no-touch areas, rollout constraints, or deadlines

If the request is vague, ask for the PR link or number and the intended review focus before proceeding.

Tooling Stance

This skill is tool agnostic.

Use the strongest available evidence from:

• PR metadata and diffs
• CI status or failing checks
• local test or typecheck commands
• migration or rollout notes
• feature-flag or config changes
• release or rollback documentation

Prefer the repository's own validation commands and only broaden if the local signals are insufficient.

Instructions

Step 1: Define The PR Scope

Capture the PR number or link, title, merge target, and changed files.

If the user gave only a branch name or short description, resolve it to the specific PR before reviewing risk.

Step 2: Map The Risk Surfaces

Inspect the diff and identify the concrete surfaces likely affected:

• APIs, handlers, jobs, or CLI entry points
• shared libraries or serialization boundaries
• schema or migration changes
• feature-flagged paths
• rollout, rollback, or compatibility-sensitive surfaces

Keep the mapping short and concrete. Do not widen the scope beyond what the PR actually touches.

Step 3: Check Validation Quality

Assess what evidence exists for the change being safe:

• tests added or updated
• existing tests that cover the changed paths
• CI status and failing checks
• local validation commands that apply to the touched surface
• missing validation for risky branches, edge cases, or migrations

Treat missing validation as a real risk signal when the PR changes behavior.

Step 4: Check Rollout And Compatibility Risk

Look specifically for:

• backward-incompatible API or data changes
• migrations that need ordering, backfills, or cleanup
• feature flags without a clear default or rollback path
• interface changes that require coordinated updates
• config changes that can fail in partially deployed states
• hidden coupling across packages, services, or consumers

If the PR depends on an assumption outside the diff, say so explicitly.

Step 5: Separate Strong From Weak Signals

Only report a finding when the evidence supports it.

Bucket observations into:

• strong risk: concrete gap, inconsistent update, or clear missing safeguard
• moderate risk: plausible issue with visible local impact
• weak signal: suspicious but not enough to justify a concrete recommendation

Do not inflate weak signals into firm findings.

Step 6: Propose The Smallest Safe Next Action

For each strong or moderate finding, recommend one small next action such as:

• add a targeted test
• document the rollout or rollback assumption
• split the change into safer phases
• add a guard or compatibility shim
• verify the migration order
• add a feature-flag default or fallback

Keep recommendations tightly scoped and reviewable.

Output Requirements

Provide a report with these sections:

1. Scope reviewed
2. Ranked risk findings
3. Weak-signal observations skipped
4. Recommended next actions

For each finding, include:

• PR number, title, and link if available
• relevant files or surfaces
• concrete risk evidence
• why this is a merge risk
• smallest safe next action
• suggested validation
• confidence level

For each weak-signal observation, include:

• what looked risky
• why the evidence was insufficient
• explicit no-action decision

Quality Bar

• Stay pre-merge and do not drift into post-merge impact analysis
• Use concrete PR and diff evidence rather than generic review language
• Treat missing validation and rollout gaps as first-class risk signals
• Keep the output short, prioritized, and directly actionable
• Prefer the smallest safe follow-up over broad redesign advice