The commit-authorship guard has a source and no caller

Split out of #17783 at the review of PR #18753 (PRR_kwDODSospM8AAAABNyLlNg, RA-1): that PR delivers #17783's R3 half, and a Resolves on a ticket still carrying R2 would have closed R2's only open home in the org. This ticket is that home. The R2 analysis below was authored on #17783 by @neo-opus-vega (Ledger row, AC) and @neo-opus-grace (the 2026-09-15 dual-caller correction) and moves here unchanged in substance.

Context

neo-agent-brain:ai/scripts/lint/check-commit-authorship.mjs refuses agent commits that carry the operator's identity or credit a co-author outside the agent roster. It was hardened by #17195 (the author-scoped boundary: an agent-authored commit admits no non-roster trailer at any domain, while an outside contributor's commit is never inspected). Nothing runs it. It has already cost once: an agent-checkout PR authored under the operator's personal account passed every gate and was caught by eye.

Measured 2026-09-16:

The Problem

The guard is already dual-caller by design, and says so in its header — "Two unforgeable sources, one per caller: Hook: checkout ownership … CI: --author-login, which the workflow fills from the GitHub-authenticated PR author. A PR cannot forge who opened it.":

function isAuthenticatedAgentLane() {
    const login = …process.argv… '--author-login';
    return isAgentCheckout() || Boolean(rosterEmailForLogin(login))
}

Two checks, two callers the guard was built for, neither wired:

The roster coupling blocks neither. The remaining work is wiring. Refuted by measurement (see Placement below): neomjs/neo cannot resolve a Brain script, and publishing the guard would publish its roster. The roster decided where the guard lives.

The Architectural Reality

• The hook half is neo-local: .husky/pre-push (set -e, payload captured once, one printf … | node … line per guard).
• The CI half is cross-repository: neo-agent-skills#14 owns the reusable PR-governance workflow each repo calls, so a per-repo workflow here is acceptable only as that ticket's recorded interim child step.
• The guard source lives in the Brain. Whether the hook calls it across a checkout boundary or it relocates beside its caller (as #17783's R3 guard did in PR #18753) is this ticket's first measured decision, not a presumption.
• Placement, decided 2026-09-19 (@neo-opus-ada's comment 5743105316, after @tobiu's input; @neo-opus-vega's measurement in 5702467350):
  • The guard's logic publishes through neo-agent-skills, which every calling repo already depends on.
  • The team roster is a caller input (--roster), never package data. It draws the line between our team and every other deployment of the Agent OS, so no package may carry it.
  • With no roster, the trailer check inspects nobody: the tenant-safe default. The operator-identity check needs no roster and runs in any deployment.
  • Our roster stays in the Brain. Agent seats pass it from their Brain runtime, and our CI callers from a pinned Brain checkout.

Contract Ledger

Retirement: fold into neo-agent-skills#14's unified governance when its reusable workflow supersedes per-repo callers.

Acceptance Criteria

• AC-1 — check-commit-authorship.mjs runs as a PR-visible check in the repo(s) whose commits it polices via the neo-agent-skills#14-sanctioned caller, with the red/green control pair from the Ledger — or its non-enforcement is recorded as an accepted risk with a named owner and a date.
• AC-2 — the hook half runs from .husky/pre-push in neomjs/neo, fed the captured payload like its siblings, with a red control from an agent checkout carrying the operator identity.
• AC-3 — .github/workflows/spec-retirement-lint.yml:22 states the pre-push guards the hook actually runs.
• AC-4 — revalidationTrigger — if neo-agent-skills#14's caller design or the guard's location changes before this closes, the body is re-measured rather than trusted.

Out of Scope

R1 (no required status context → #17171 + neo-agent-skills#14) · R3 (the parity guard → #17783 / PR #18753) · R4 (→ neo-agent-brain#250) · rewriting the guard's checks (the roster becomes an input; what the checks refuse stays).

Related

#17783 (source ticket) · neo-agent-skills#14 (caller authority) · #17195 (the author-scoped boundary this guard enforces) · #16143 · #16280

Live latest-open sweep: checked the latest 20 open issues at 2026-09-16T08:04Z; no equivalent. Cross-org search for check-commit-authorship across neo, neo-agent-brain, neo-agent-skills, neo-agent-institution: only #17783 and neo-agent-skills#14 are open, and #14 carries no AC for this guard. A2A claim sweep: no claim on this scope. Memory Core rationale sweep: #17195's design history; nothing contradicts a dedicated caller ticket.

Owner: @neo-opus-ada (claimed 2026-09-19, after the operator's "it absolutely matters").

Origin Session ID: a758f0e1-0dc1-4e4e-a975-4824a9319487

Retrieval Hint: "commit authorship guard caller pre-push hook author-login CI Skills 14"