kc-pr-review-resolve

All text output follows unified language preference. See plugin CLAUDE.md for query flow.

REQUIRED SUB-SKILL: Use superpowers:receiving-code-review for evaluation mindset — verify before implementing, push back on incorrect suggestions, no performative agreement. This skill handles mechanics; that skill governs judgment.

Configuration

This skill respects pr_review_resolve.auto_confirm in the adopter's workflow config (project CLAUDE.md or workflow README):

Rationale: reply_only mode addresses the daily-pain segment "captain must confirm action plan even when all findings are nits/FP" without granting the skill authority to push controversial fixes. Any issue requiring code change still gates on captain confirmation.

Future extension (separate entity): trivial_fix mode for single-line typo / null-check / unused-import auto-fixes. Out of scope for this revision.

Process Flow

digraph resolve_pr {
  rankdir=TB;
  node [shape=box];

  detect [label="Detect PR\n(from args or current branch)"];
  fetch [label="Fetch unresolved threads\n+ PR-level reviews\n+ reviewer metadata"];
  validate [label="Dispatch review agents\nto validate each thread"];
  dedup [label="Cross-AI dedup\ngroup by (file, conceptual issue)"];
  verdict_dedup [label="Cross-review verdict dedup\n(suppress prior-dismissed\nunless file changed)"];
  triage [label="Triage & classify threads\n(AI reviewer annotations)"];
  report [label="Report findings to user"];

  node [shape=diamond];
  auto_confirm_check [label="Step 4.5:\nauto_confirm config?\n(reply_only: skip-fix if\nconditions met;\npreapproved: fix if\ndirective given)"];
  confirm [label="User confirms\naction plan?"];

  node [shape=box];
  fix [label="Fix valid issues\n(one commit per fix)"];
  push [label="Push changes"];
  reply [label="Reply to ALL threads\non PR"];
  resolve_threads [label="Resolve fixed threads"];
  rereview [label="Smart re-review\n(AI-aware tagging)"];

  detect -> fetch -> validate -> dedup -> verdict_dedup -> triage -> report -> auto_confirm_check;
  auto_confirm_check -> confirm [label="off / blocked"];
  auto_confirm_check -> reply [label="reply_only\nauto-confirmed\n(skip fix)"];
  auto_confirm_check -> fix [label="preapproved\nauto-confirmed\n(directive detected)"];
  confirm -> fix [label="yes"];
  confirm -> report [label="no, adjust"];
  fix -> push -> reply -> resolve_threads -> rereview;
  reply -> resolve_threads;

  node [shape=diamond];
  has_ai [label="AI reviewer\nre-triggered?"];

  node [shape=box];
  offer_monitor [label="Offer to monitor\nAI response"];
  monitor [label="Poll for new\nAI feedback"];
  report_new [label="Report new feedback\n(loop back to triage)"];

  learning [label="9: Learning\nskill + project\nknowledge capture"];

  node [shape=doublecircle];
  end_node [label="Done"];

  rereview -> has_ai;
  has_ai -> offer_monitor [label="yes"];
  has_ai -> learning [label="no"];

  offer_monitor -> monitor [label="user accepts"];
  offer_monitor -> learning [label="user declines"];
  monitor -> report_new [label="new feedback"];
  monitor -> learning [label="timeout / no response"];
  report_new -> triage [label="re-triage"];
  learning -> end_node;
}