Add role-first lane pickup before review intake

Context

@tobiu surfaced a workflow friction during GPT harness recovery on 2026-05-19: this is not a blame or under-target quota problem. The structural issue is trigger ordering. If an agent starts from a fresh boot, watchdog wake, or PR-review request, it can enter reviewer mode before the role/lane-pickup substrate fires. The existing CI-wait routing is useful after a PR exists, but it does not solve the earlier state where no author lane has been picked yet.

V-B-A performed before filing:

• Mailbox checked: 0 unread at the time of ticket drafting.
• Current role payloads already encode the desired behavior: lead-role-mode.md says lead means pick own lane visibly + focus + V-B-A + challenge (.agents/skills/lead-role/references/lead-role-mode.md:7-13) and peer-role-mode.md says peer means proactive V-B-A/challenge, with self-selection instead of waiting for assignment (.agents/skills/peer-role/references/peer-role-mode.md:5-18, :46-48).
• post-review-pickup/SKILL.md only invokes after review/review-response handoff (.agents/skills/post-review-pickup/SKILL.md:1-8). Its deeper workflow contains a hidden lane-discovery trigger for fresh-boot (.agents/skills/post-review-pickup/references/post-review-pickup-workflow.md:56-60), but the router does not expose that trigger, so the payload is not guaranteed to be read before the first review.
• fair-band-pre-flight-gate.md explicitly notes lane-discovery discipline is skill-invocation-dependent and that the PR-open gate only catches paths that reach PR authoring (.agents/skills/pull-request/references/fair-band-pre-flight-gate.md:1-4).
• Live distribution check over the last 200 merged PRs: neo-gemini-pro=83, neo-opus-ada=68, neo-gpt=45, dependabot[bot]=4. Live last-30 check was effectively balanced: Gemini=10, GPT=10, Claude=9, dependabot=1. That means the ticket should not be framed as immediate quota correction; it is a decay-prevention trigger-ordering fix.

The Problem

The current substrate solves the author-after-review problem better than the pre-review intake problem:

1. After a review is posted, post-review-pickup can require a next lane or explicit halt.
2. After a PR is opened, the FAIR-band PR-body gate can enforce author-lane stance.
3. During lead-role or peer-role, the payloads already tell agents to self-select lanes.

But a fresh session can begin with live review requests and never cross the earlier lane-pickup surface. The agent can perform reviews, send review A2A, then continue to the next review because every local trigger points at review lifecycle work. The result is a reviewer-only cycle even though the correct mental model is: pick a lane, create a PR, then review peer PRs afterwards or while own CI is pending.

This is a trigger-ordering hole, not an individual performance target.

The Architectural Reality

The relevant surfaces are agent-skill routers and payload maps, not runtime framework code:

• /lead-role already has the correct lane-selection primitive, but it only fires when lead-role is explicitly invoked or a baton/operator trigger is present.
• /peer-role already says self-select lanes and resist waiting for assignment, but it is framed around substrate validation and convergence work, not ordinary review-queue intake.
• /post-review-pickup contains the deeper lane-discovery hook, including fresh-boot, but its SKILL.md trigger is post-review only.
• /pr-review is intentionally concrete-PR-focused. Adding a giant lane-management manual there would bloat the hot review path and violate progressive-disclosure discipline.
• create-skill guidance requires lightweight routers and payload extraction; any fix must use a small map trigger and avoid stuffing workflow rationale into SKILL.md.

The Fix

Add a minimal, progressive-disclosure pre-review intake gate for agents that are about to enter review work with no active author lane in the current session.

Candidate shape:

• Add a small router-visible trigger to the appropriate skill surface, likely post-review-pickup/SKILL.md or a narrow payload referenced from pr-review/SKILL.md, that fires before first review intake from fresh boot / watchdog / review-wake when no current author lane exists.
• The gate should require a quick lane-discovery check: open self-pickable tickets / assigned-to-me / recent operator focus / current public authority, then either pick an author lane or document why review-first is the positive-ROI move.
• If role context is active or explicitly named, read /lead-role or /peer-role before lane choice, not after review work.
• Preserve the CI-wait routing: after opening a PR, peer review work is appropriate while own CI is pending, with a return-to-own-PR check.
• Keep top-level router changes tiny. Put rationale, examples, and exceptions in a payload file.

Contract Ledger Matrix

Acceptance Criteria

• A fresh-session or watchdog review-intake path has a visible trigger that runs lane discovery before the first PR review when no current author lane exists.
• The implementation preserves non-blame language: no quotas, no per-agent shaming, no forced assignments.
• The gate distinguishes role active from ordinary review intake; /lead-role and /peer-role are read first only when their real triggers apply.
• The gate includes explicit exceptions for urgent/human-directed review and for review work that is intentionally done while the agent's own PR is waiting on CI.
• The fix follows progressive-disclosure discipline: minimal router text, detailed rationale in payload, no broad AGENTS.md expansion unless V-B-A proves a hot-path anchor is required.
• Review evidence includes a fresh simulation or written walkthrough of: boot -> review request present -> no author lane -> lane discovery -> either author lane picked or review-first rationale recorded.
• If .agents/skills/skills.manifest.json is affected, node ai/scripts/lint-skill-manifest.mjs --base origin/dev passes.

Out of Scope

• No change to human-only merge authority.
• No ranking scoreboard or author-count enforcement.
• No requirement that every review wait for an author PR.
• No new role hierarchy; maintain the flat peer-team model.
• No Body-layer runtime changes.

Avoided Traps

• Quota framing: the last-30 window is balanced, so this is not an immediate under-target correction. The long-window imbalance is only evidence that the decay mode matters.
• Forcing role-mode everywhere: lead-role and peer-role are not generic wrappers around every review. The fix should make role payloads fire when appropriate, not create synthetic role state.
• Stuffing pr-review/SKILL.md: review template loading is already heavy. The right shape is a compact trigger plus payload pointer.
• Reviewer freeze: peer review remains valuable, especially while own CI is pending. The fix is to pick/justify lane order, not to block review work.

Related

• #11430 closed adjacent work: FAIR-band author-lane pickup discipline after PR review/response handoff.
• #11603 broad adjacent friction: substrate bandwidth and contribution distribution counter-bias.
• #11605 broad adjacent friction: skill baggage / decision-atom cleanup.
• Discussion #11206 / #11209 lineage via lead-role and peer-role coordination payloads.

Handoff Retrieval Hints

• Query Memory Core for: role-first lane pickup before review intake, reviewer-only cycle fresh boot, post-review-pickup fresh-boot trigger hidden, lead-role peer-role lane self-select before review.
• Verify current facts with: gh search prs --merged --repo neomjs/neo --limit 200 --sort updated --json author --jq 'map(.author.login) | group_by(.) | map({author: .[0], count: length}) | sort_by(.author)'.