Skip to content

perf(implementation): drive one PR with one owner from create to merge instead of per-phase agents #8754

Description

@kyle-sexton

No related issue: found while landing #8714.

Problem

#8714 went from creation to merge in about 3 hours, while each CI-plus-review round took about 10 minutes. Most of the gap came from handing the PR between separate agents for create, ready and monitor. Each agent waited for its own background checks before reporting, and only then did the next step start, so every handoff added idle time.

Fix

The orchestrating skills (/implementation:implement-dispatch Step 5, /work-items:work) drive one PR with one long-lived owner from create to merge, using /source-control:pull-request full or a single monitor loop. They do not dispatch a fresh agent for each phase. A handoff that does happen carries an event-driven wait (a Monitor watch or push channel), not a poll.

Verification

A dispatched run's PR shows no gap longer than one CI round between ready and the first monitor action.

Related

#8714

Activity

  1. added
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Oct 21, 2026
  2. github-actions commented on Oct 21, 2026

    @github-actions
    Contributor

    Proposed work class: work-class: scoped
    Basis: automated intake triage; the lanes App applied this class.

    Size: Small
    Complexity: Low
    Uncertainty: Known
    Flags: none
    Blocked by: none

    Decision comment from kyle-sexton: #8754 (comment)

    Decision recorded. The owner answered in the Refinement and Issue Triage sessions (2026-10-25), deciding the open items below on the facts:

    go with whatever the facts/data/metrics + /discovery:research says and ensure we're aligned with SOLID, DRY/SSOT principles

    1. Which skill owns the change? Decided: /work-items:work only.
    2. Does the one owner now merge? Decided: No. Merging stays with the babysit lane or a person.
    3. What does the one owner replace? Decided: The separate monitor and post-green review dispatches, as one strong-tier agent.
      [quote truncated; the full decision is at the link]

    The decision settles all four open questions, so I'm asking nothing further. The scope is a bounded edit to the /work-items:work skill: one strong-tier owner dispatched as soon as create --pushed returns, event-driven waits, fresh fix workers, and the never-merge boundary unchanged. The live-PR timing check needs a real dispatched run, so a person has to do it.

    Automated triage, applied by the lanes App.

  3. added and removed
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Oct 21, 2026
  4. github-actions commented on Oct 26, 2026

    @github-actions
    Contributor

    The triage sweep dispatched intake triage for this issue (attempt 1 of 2).

  5. added
    status: readyTriaged, unblocked, and fully specified; eligible to pick up.
    needs-owner-decisionAwaiting a human or maintainer judgment call.
    and removed
    status: readyTriaged, unblocked, and fully specified; eligible to pick up.
    on Oct 26, 2026
  6. melodic-automation-lanes commented on Oct 26, 2026

    @melodic-automation-lanes
    Contributor

    Not admitted to the issue lane: no trusted-human acceptance criteria: a listed human must put the confirmed - [ ] criteria in the body or post them as their own comment, or, on an issue a listed human filed, intake's newest refinement comment must hold them unedited. Admission takes the issue once each of those is met, when intake triages it again (.github/actions/resolve-config/admission-rules.json).

  7. melodic-automation-lanes commented on Oct 26, 2026

    @melodic-automation-lanes
    Contributor

    The latest triage verdict holds nothing: the issue is at status: ready.

  8. melodic-automation-lanes commented on Oct 26, 2026

    @melodic-automation-lanes
    Contributor

    Refinement. A read-only run explored this repository, searched its issues and pull requests, and read the allowed docs to draft what follows. On an issue a listed human filed, it counts as the issue's spec once posted, while unedited; it is still model text, so check it.

    Intended outcome: The PR-driving steps in work-items work and the implement-dispatch handoff keep one long-lived owner on the PR from create to merge-ready, and any remaining handoff waits on an event instead of polling.

    Related work

    Open questions

    1. The issue names implement-dispatch Step 5, but that skill has no such step. It hands off to /implementation:implement Step 5, which stops at the pre-PR sequence. Which skill should own driving the PR?
      RECOMMENDED: Put the change in work-items work (the orchestrator-owned PR step and the post-green review step) and in the implement-dispatch handoff row, and drop the 'Step 5' wording. Basis: .base/plugins/implementation/skills/implement-dispatch/SKILL.md:283
    2. The issue says 'create to merge', but work ends at PR creation and handoff to babysit-prs, and its never-merge boundary says merging belongs to the babysit lane or a person. Should the single owner now merge, or stop at merge-ready?
      RECOMMENDED: Keep the never-merge boundary: one owner drives create, ready and monitor to green, then hands off to babysit-prs. Read 'to merge' as 'to merge-ready'. Basis: .base/plugins/work-items/skills/work/SKILL.md:317
    3. work dispatches monitor and post-green review as separate agents into the persisted worktree at different model tiers (fast for monitor, strong for review), plus fresh fix workers each round. Should one owner replace all of these, or only the create, ready and monitor handoffs?
      RECOMMENDED: Replace only the create, ready and monitor handoffs with one owner. Keep fresh scope-fenced fix workers and the strong-tier review, since the worktree carries state and a reviewer must not be weaker than the implementer. Basis: .base/plugins/work-items/skills/work/SKILL.md:313
    4. The Verification line ('no gap longer than one CI round between ready and the first monitor action') can only be checked on a live dispatched PR. What should an unattended run check instead?
      RECOMMENDED: Check that the skill text names one owner, forbids a per-phase dispatch, and requires an event-driven wait (Monitor or push channel) on any handoff, plus an eval in evals.json asserting it. Leave the live-PR timing as a person's check. Basis: judgment

    Sources read

    • /home/runner/work/_temp/refine-data/context.json: issue 8754 has no task list; label status: needs-decision; no maintainer decision comments
    • plugins/implementation/skills/implement-dispatch/SKILL.md:283: the all-phases-complete row hands off to /implementation:implement Step 5 and does not open the PR
    • plugins/implementation/skills/implement/SKILL.md:280: Step 5 hands off to the pre-PR sequence, ending in /source-control:pull-request
    • plugins/work-items/skills/work/SKILL.md:313: orchestrator runs create --pushed, then dispatches monitor and post-green review separately into the persisted worktree, and fresh fix workers
    • plugins/work-items/skills/work/SKILL.md:315: post-green review, then handoff to /source-control:babysit-prs
    • plugins/work-items/skills/work/SKILL.md:317: never-merge boundary: work ends at PR creation and handoff

    Automated refinement by the lanes App. Answer with a comment whose first line is **Decision recorded.**.

  9. kyle-sexton commented on Oct 26, 2026

    @kyle-sexton
    MemberAuthor

    Decision recorded. The owner answered in the Refinement and Issue Triage sessions (2026-10-25), deciding the open items below on the facts:

    go with whatever the facts/data/metrics + /discovery:research says and ensure we're aligned with SOLID, DRY/SSOT principles

    I should only ever be in the loop ONLY when necessary (intent/vision changed, something came up during implementation that impacts trajectory/alignment, OR a high security risk) - I've covered this all before

    No owner ledger (.work/*/interview-checklist.md, .work/gh-monitor-owner-decisions/) names this issue. Read against origin/main at a6ba1c4c18.

    Part of the body is on main since #8714. /work-items:work now has create --pushed ready the PR itself, "so the orchestrator dispatches no ready of its own" (plugins/work-items/skills/work/SKILL.md:313; added by d25b18a, #9483, 2026-10-24). So the create-to-ready handoff is gone. What is left: after the PR is open, the orchestrator dispatches the monitor watch and the post-green review pass as separate agents into the persisted worktree, the monitor at the fast tier and the review at the strong tier (:313), then hands off to /source-control:babysit-prs (:315).

    1. Which skill owns the change?
      Decided: /work-items:work only. /implementation:implement-dispatch dispatches no PR phase: its last row hands to /implementation:implement Step 5 (plugins/implementation/skills/implement-dispatch/SKILL.md:283), which hands the PR to the pre-PR sequence in the same session (plugins/implementation/skills/implement/SKILL.md:280).
    2. Does the one owner now merge?
      Decided: No. Merging stays with the babysit lane or a person (work/SKILL.md:317). That boundary is a merge gate, and this record does not weaken a gate. The body's measure, the gap between ready and the first monitor action, needs no merge by work.
    3. What does the one owner replace?
      Decided: The separate monitor and post-green review dispatches. One agent, dispatched into the persisted worktree at the strong tier as soon as create --pushed returns, monitors to green and runs the review pass. The strong tier is required because the review must be no weaker than the implementer (:313); the monitor's fast tier is a "may", not a rule. Fix workers stay fresh scope-fenced dispatches, because the worktree, not the agent, carries state across them (:313).
    4. How does a handoff wait?
      Decided: With the existing rules, no new mechanism. Monitor's Phase 3 picks a push channel first, then its fallback (plugins/source-control/skills/pull-request/SKILL.md:151, Step 2), and dispatched workers wait with run_in_background or Monitor, never a poll loop (implement-dispatch/SKILL.md:302). The skill text points at these.

    Acceptance criteria:

    • plugins/work-items/skills/work/SKILL.md dispatches one agent into the persisted worktree, at the strong tier, as soon as create --pushed returns, and that agent runs both the monitor watch and the post-green review pass. No separate monitor dispatch remains.
    • The same step says the orchestrator does not wait on CI before that dispatch, and that any wait on the owner is event-driven per the monitor phase's push-channel rule, never a poll.
    • Fix workers are still fresh dispatches into the persisted worktree, and the never-merge boundary and babysit handoff are unchanged.

    Host run (host-run label):

    • On a dispatched run's PR, no gap longer than one CI round between ready and the first monitor action (the body's Verification).
  10. added
    admission-deferredAdmitted but waiting for WIP room; the hourly work-unblock sweep labels it agent-ready.
    on Oct 28, 2026
  11. added
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    and removed
    admission-deferredAdmitted but waiting for WIP room; the hourly work-unblock sweep labels it agent-ready.
    on Oct 28, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.status: readyTriaged, unblocked, and fully specified; eligible to pick up.

    Type

    Fields

    No fields configured for Task.

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions