feat(cos): multiple ordered code-review agents per task

[atomantic]

Summary

• Replaces the single per-task reviewer with an ordered reviewers list, plus reviewStopMode (all/on-findings/on-clean) and reviewerApplies, so the custom-task form and scheduled-task form can run codex/gemini/copilot in a chosen order.
• New ReviewerPicker component (click-order pills with ↑↓ reorder + ✕ remove, stop-mode select shown for 2+ reviewers, reviewer-applies toggle shown when a non-copilot reviewer is present). Wired into TaskAddForm, ScheduleTab, and the AgentCard badge.
• Threads the list end-to-end: schema + sanitizeTaskMetadata, cos.addTask, both spawn paths' cleanup (agentLifecycle runner path and agentCliSpawning direct path), and the agent prompts as slashdo --review-with a,b,c [--review-stop-on-findings|--review-stop-on-clean] [--reviewer-applies]. The system-spawned review-loop follow-up runs each reviewer in order; Copilot is pre-requested only when it's in the list, and CLI reviewers still run on non-GitHub forges.
• Read-side normalizeReviewers() (server validation.js + client constants.js mirror) keeps legacy single-reviewer tasks/schedules working.
• Bumps the bundled lib/slashdo submodule v2.18.0 → v3.1.0 (adds the multi-reviewer --review-with list + stop-mode + --reviewer-applies), and adds .claude/commands/do/{depfree,pr-better,scan}.md symlinks to match the new command set.

Test plan

• cd server && npm test — full suite green (incl. new validation, agentPromptBuilder, cleanupAgentWorktree, agentCliSpawning reviewer cases)
• cd client && npm test — green (incl. new ReviewerPicker tests)
• npm run build — clean
• Browser: enabled Review Loop in the task form, added/reordered/removed reviewers, confirmed stop-mode + reviewer-applies appear conditionally
• Multi-reviewer review loop (codex → gemini → copilot) runs on this PR

[Copilot]

Pull request overview

This PR upgrades the CoS “review loop” from a single per-task reviewer to an ordered multi-reviewer pipeline (e.g., codex → gemini → copilot), with configurable stop behavior and an option for CLI reviewers to apply fixes automatically. It threads the new reviewer list through task creation, validation/sanitization, agent spawning/cleanup (runner + direct CLI spawn paths), prompt generation, and the client UI (task + schedule forms, agent card badge), while keeping legacy reviewer tasks working via normalizeReviewers().

Changes:

• Replace singular reviewer with ordered reviewers[], plus reviewStopMode and reviewerApplies, end-to-end (schemas, metadata sanitization, addTask, spawning/cleanup, prompts).
• Add ReviewerPicker UI (add in click-order, reorder, remove; conditional stop-mode + reviewer-applies controls) and wire it into TaskAddForm and ScheduleTab.
• Update prompt instructions + tests to reflect multi-reviewer slashdo flags (--review-with a,b,c, --review-stop-on-*, --reviewer-applies) and multi-reviewer follow-up behavior.

Reviewed changes

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
server/services/taskSchedule.js	Updates task prompt wording to describe ordered multi-reviewer review steps.
server/services/cos.js	Threads normalized reviewer lists into task prompt templating and task metadata construction.
server/services/cleanupAgentWorktree.test.js	Expands coverage for multi-reviewer follow-up spawning and Copilot pre-request rules.
server/services/agentPromptBuilder.test.js	Updates/extends prompt expectations for reviewer lists and new flags/stop-modes.
server/services/agentPromptBuilder.js	Implements multi-reviewer prompt wiring, flag construction, and updated completion workflow text.
server/services/agentLifecycle.js	Updates spawn/cleanup paths to pass reviewer lists and handle Copilot pre-request only when leading.
server/services/agentCliSpawning.test.js	Ensures direct CLI spawning path passes reviewers/stop-mode/applies into cleanup.
server/services/agentCliSpawning.js	Threads normalized reviewers + flags through direct-spawn cleanup options.
server/lib/validation.test.js	Adds tests for reviewers list normalization and slashdo arg building.
server/lib/validation.js	Adds normalizeReviewers, stop-mode constants, and buildReviewWithArgs; extends schemas/sanitization.
client/src/components/cos/TaskAddForm.jsx	Replaces reviewer dropdown with ReviewerPicker and submits new reviewer metadata fields.
client/src/components/cos/tabs/ScheduleTab.jsx	Replaces reviewer dropdown with ReviewerPicker for scheduled tasks and writes new metadata keys.
client/src/components/cos/tabs/AgentCard.jsx	Updates review-loop badge to display ordered reviewer lists.
client/src/components/cos/ReviewerPicker.test.jsx	Adds unit tests for add/remove/reorder and conditional controls.
client/src/components/cos/ReviewerPicker.jsx	New ordered multi-reviewer picker component.
client/src/components/cos/constants.js	Adds client-side reviewer normalization + stop-mode constants (mirroring server).
.changelog/NEXT.md	Documents the new multi-reviewer capability and slashdo submodule bump.

Comments suppressed due to low confidence (1)

server/services/agentLifecycle.js:1547

• The documentation comment above spawnReviewLoopFollowUp still describes spawning a Copilot-only /do:rpr loop and GitHub-only behavior. The implementation now supports an ordered reviewers list (dropping copilot only on non-GitHub forges) plus reviewStopMode/reviewerApplies. Please update that docblock to match the new multi-reviewer semantics and parameter names.

 * branch (via createWorktree's `existingBranch` option) so it can fix-and-push
 * without trampling concurrent agents.
 */
export async function spawnReviewLoopFollowUp({ originalAgentId, originalTask, prUrl, prBranch, sourceWorkspace, reviewers = DEFAULT_REVIEWERS, reviewStopMode = DEFAULT_REVIEW_STOP_MODE, reviewerApplies = false }) {
  if (!prUrl || !prBranch) return null;

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

server/services/agentPromptBuilder.js:286

• The follow-up merge/verification steps are hardcoded to gh pr merge and gh pr view, but follow-up tasks may be spawned for GitLab MRs when CLI reviewers are configured (copilot is stripped, not the whole follow-up). On non-GitHub forges these commands will fail, leaving the MR open and the worktree/branch potentially leaking. Please choose the correct forge CLI (gh vs glab) for merge/view based on reviewLoopPRHost/reviewLoopPRUrl (or, if review-loop follow-ups are intended to be GitHub-only, block spawning on non-GitHub hosts even for CLI reviewers and emit a clear warning).

   gh pr merge "${prUrl}" --squash --delete-branch
   \`\`\`
   ${prOwner && prRepo && prNumber ? `(Equivalent: \`gh pr merge ${prNumber} --repo ${prOwner}/${prRepo} --squash --delete-branch\`.)` : ''}
   You have already verified the review is clean, so force the immediate merge. Adding any merge-deferral flag would leave the PR open after you exit.
5. Confirm the PR is actually merged before exiting: \`gh pr view "${prUrl}" --json state -q .state\` must return \`MERGED\`. If it returns \`OPEN\` or \`CLOSED\`, investigate (a check is failing, a thread is still unresolved, or branch protection is blocking) — fix and retry the merge. Do NOT exit until state is \`MERGED\`.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

server/services/agentLifecycle.js:1597

• This metadata block includes a comment about skipCommitCheck, but no such metadata field is actually set here (and there are no other references in the codebase). This is misleading for future maintainers—either add the intended skipCommitCheck flag (if it exists) or remove/update the comment to match actual behavior.

      sourceTaskId: originalTask?.id || null,
      sourceAgentId: originalAgentId || null,
      // skipCommitCheck: this task may exit cleanly with zero new commits if the
      // initial Copilot review came back clean. Don't false-positive that as failure.
      readOnly: false

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

server/services/agentLifecycle.js:1485

• canSpawnFollowUp is computed from the full normalized reviewer list, but spawnReviewLoopFollowUp() may drop copilot on non-GitHub forges and return null when that empties the effective list. Right now that results in silently not spawning any follow-up (and no warning) even though reviewLoop was enabled. Consider detecting a null return here and emitting a log/warning explaining that the reviewer chain is empty after forge filtering (e.g., copilot-only on GitLab), so the PR/MR won’t be auto-reviewed/merged.

      const canSpawnFollowUp = shouldRequestCopilot && reviewerList.length > 0;
      if (canSpawnFollowUp) {
        await spawnReviewLoopFollowUp({
          originalAgentId: agentId,
          originalTask,

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated 2 comments.

[Copilot]

Pull request overview

Copilot reviewed 25 out of 25 changed files in this pull request and generated 3 comments.