LoopX PR Review

LoopX PR Review

This skill is a thin host adapter. The built-in pull-request-review capability owns review depth, evidence requirements, completeness, and verdict policy through the CLI packet. Do not copy those rules into this skill or replace them with a host-specific checklist.

Route

Use this skill for /loopx-pr-review, explicit PR reviews, or review queues by state or time window. Route approval, merge, self-merge, and admin bypass to loopx-pr-merge (optional repo-kept workflow, not installed by default) after the evidence review is complete; it never replaces this skill's exact-head gate.

For named PRs, resolve heads and run repeatable --target-exact-head NUMBER@HEAD_OID. An omitted --state keeps ordinary queue discovery open-only while exact targets remain lifecycle-neutral; use explicit --state merged|all only for deliberate history or post-merge audit.

Translate only explicit filters:

• --repo owner/repo
• --since ISO
• --state open|merged|all
• --limit N; --target-exact-head NUMBER@HEAD_OID (repeatable direct read)
• --review-priority other-developers-first|owner-first (default other-developers-first; use owner-first to opt into owner priority) When omitted, the CLI resolves pull_request_review from the standard machine capability editor; an absent namespace keeps the default other-developers-first. Words such as today, open, or merged are filters, not permission to return a table only; stats-only output needs an explicit opt-out such as 只统计 or stats only.

Preserve The Packet

Save the full first JSON packet before printing a compact projection. Keep all paths named by agent_response_contract.required_packet_fields_to_preserve:

• agent_response_contract.review_execution_contract
• result_completeness, scheduling_policy, review_groups
• pull_requests[review_action_kind!=null].review_plan
• pull_requests[review_action_kind!=null].review_template
• pull_requests[review_action_kind!=null].evidence_commands

Do not pipe the only copy through jq. When an exhaustive queue request has result_completeness.complete=false, rerun with its recommended_limit before reviewing; limit_scope=exact_targets is already complete for the named targets.

Read review_execution_contract.policy_revision from the packet; require the result's review_policy_revision to equal it, never a literal this file pins. If missing or unequal, do not publish APPROVE; a conservative REQUEST_CHANGES is allowed only when it names the incompatible-policy gap. Do not retain expired temporary worktree overrides. Report incompatible policy rather than downgrading the review.

Read The Review Frame First

Read the PR's existing comments and cited documents first: a maintainer comment names the contract the change is judged against.

• Review inside that frame; cite the document and check its rows, gates, or exit criteria one by one at the exact head, attaching each finding to its row.
• Name unsatisfied rows as missing frame items with their own evidence; keep the declared boundary (preview versus promotion, milestone entry versus merge).
• Publish the frame-aligned conclusion as a PR comment citing that guidance, since a published review body cannot be edited.

Execute One Review Plan

Receive pending current-session requests through review_execution_contract.decision_procedure.establish_goal before generic queue selection, including other agents sharing a GitHub account; task intake and claimed action authority are separate judgments. Resolve the selected requests with repeatable --target-exact-head NUMBER@HEAD_OID. Follow scheduling_policy and its ranked actionable review_sequence; explicit current-request PR selection may override ordering only, never pull_requests[].review_action_kind or exact-head idempotency. Generic re-review, 重新review, and 复审 wording selects the named PR; it is not a force-refresh token. Todo/monitor prose may not select work. When review_action_kind is null, the row stays in pull_requests inventory but must not appear in review_sequence; its review_plan and review_template are null and evidence_commands is empty. Do one compact exact-head conclusion readback and report the existing verdict or bounded invalid/missing reason. Run a fresh audit only when the user explicitly requests fresh evidence despite that result, or supplies a concrete new concern; regenerate with --fresh-audit-exact-head NUMBER@HEAD_OID, then execute the complete current plan and never inherit the earlier approval. For every actionable PR:

1. Record the packet's exact head, then follow review_execution_contract.decision_procedure: the current goal and problem_context delivery judgment first, including on re-review, then evidence_commands and repository-native validation.

2. Fill review_plan.result_template; preserve missing evidence as unverified and never infer verified from metadata or CI. Execute its repository-reuse, default-off, authority and real-path counterfactuals rather than repeating them as prose. Fill result.reviewer per review_execution_contract.reviewer_declaration, open the body with its body_marker line, and read problem_context.spec_basis's specification before the diff.

3. Apply completion_gate literally: save final Markdown in review_body, then check evidence and that exact body. Follow capability-owned floors and scope counterfactuals; prose cannot replace missing execution:

   loopx --format json pr-review --check-result review-result.json --packet review-packet.json

   Fix contradictory verdicts, not evidence labels. It cannot verify evidence truth, architecture, freshness or a relabeled old result.

4. Publish the checked review_body; recheck after edits. Replace result.review_body with the remote readback and rerun --check-result; matching does not certify reasoning.

5. Re-read the remote head immediately before verdict and publication; restart evidence if it changed.

Each PR needs independent evidence and a standalone card; a queue table is a preface only.

For managed review, pass --goal-id GOAL and follow the packet’s resolved wait_for_ci: false means never fetch, poll, or wait for CI; true retains available CI observation and the merge gate's CI policy. It does not require every CI job to finish or succeed before APPROVE when independent current evidence covers the changed invariants. Follow validation_matrix.validation_source: required marks decisive review evidence, not branch protection; record merely pending remote jobs as separate diagnostic rows, and keep an invariant unverified when CI is its only decisive coverage. Pending CI alone is never a REQUEST_CHANGES reason. Apply validation_matrix.failure_attribution, including its evidence_scope, to current required checks. An independently attributed unchanged baseline failure or external outage can hold merge readiness without forcing REQUEST_CHANGES on an unrelated PR. Preserve earlier failures and the current evidence that supersedes them in existing evidence fields; historical root-cause completeness alone is not an approval gate. Current unattributed failures, material instability and missing affected-invariant coverage still block; selecting one successful rerun does not resolve them. Configure one Goal with configure-goal --goal-id GOAL --no-pr-review-wait-for-ci --execute; clear with --clear-pr-review-configuration --execute.