docs: record post-M7 sequencing decisions and authorize wave 1

[aryeko]

What

Docs-only PR that makes MILESTONES.md (plus one small ROADMAP.md touch) the durable
record of the owner's 2026-07-04 post-M7 sequencing decisions, and proposes a new M8
milestone for owner ratification in this review.

This PR is human-reviewed and human-merged. It is explicitly excluded from the wave-1
autonomous-squash-merge authorization it itself records (decision 5) — do not enable
auto-merge, do not merge without explicit human sign-off.

Where it lands

• MILESTONES.md — a dated "Post-spine decisions (2026-07-04)" subsection appended under
  M7 (after the existing "Closeout checkpoint", before "Repo planning handoff"); existing
  M7 text is untouched, this only appends. Plus a new ### M8: Planning-discipline uplift
  entry using the file's own Milestone Record Format, State: proposed.
• ROADMAP.md — only the jig layer's "Next step" paragraph, updated in place (5 lines).
• No other files touched; profile/ and templates untouched.

The five decisions recorded (plus the wave-1 authorization)

1. D-EXIT (M7 exit evidence) — Option A. A committed, redacted EVRUN record is mandatory
   for M7 exit. EVRUN-partial (scripted injected agent leg, explicit Limitations section) is
   sufficient for exit; EVRUN-full (Codex-driven) is a named debt retired by the post-M7
   Codex-transport track (N1). M7's State stays current, flips to done only when
   EVRUN-partial merges. Concrete transport evidence gates the v0 freeze, not org-level exit.
   Lifecycle-edge split: no codex-plugin-cc-style lifecycle edges (broker, interrupt, cleanup,
   Windows process-tree) gate M7 exit — all of them gate the transport ADR and, through it,
   the freeze.
2. Packaging (N2) — internal SDK boundary now, no publishing. Packages stay private: true; no stability promise now; posture can flip to public later. The third-party
   provider-ecosystem question stays explicitly open. jig's SDK-boundary design cycle (N3) is
   authorized to start once this is recorded in jig's product docs.
3. TUI/dashboard — extension-point-only (jig guarantee CFG-7). No first-party TUI cycle;
   reopen only on recorded operator pain.
4. Smoke target — hybrid. A local bare git remote for repeatable push-path smoke tests
   (no credentials) + a dedicated private sandbox repo (jig-smoke-target, repo-scoped
   fine-grained PAT, env-only) for full landing evidence (push + PR + merge + block
   surfacing). An independent secret-scan is required on any evidence record before commit.
5. Wave-1 merge policy. Autonomous squash-merge authorized for wave-1 PRs once review
   threads resolve and pnpm check is green — except this PR.

Wave-1 authorization (recorded as decision 6 in the new subsection): jig's
absorption-ledger absorbed-vs-open update; jig's Codex-transport evidence-capture plan plus
a dated/hashed evidence/ convention; jig's product amendment recording decision 2; a
design-flow item for self-report-only/mock-adequacy conformance vocabulary; the
EVRUN-partial evidence task (after the smoke target exists); and M8 below, as a proposal
pending this review's ratification. M6 (Learning Loop Seed) stays proposed; pull timing
remains an open org choice.

M8 proposal (for ratification in this review)

### M8: Planning-discipline uplift — primary owner design-to-plan, participating
technical-design + jig. Ports the prev-gen authoring-standard's durable planning-quality
ideas, de-productized rather than ported wholesale. State: proposed; exit criteria and
artifacts intentionally left undefined pending planning, per the file's Planning Rules
(exit criteria are set when a milestone is planned, not invented ahead of it). Carve-out:
jig's PlanValidator plan-ingest preflights may proceed jig-locally under M4's existing
plan-validation-feedback ownership without waiting for M8.

This entry is a proposal, not a fait accompli — flag in review if the owner does not want
to ratify M8 as scoped, and I will amend or drop it.

Repo-plan open-question dispositions (jig's docs/delivery/m7-real-providers/repo-plan-m7.md)

Each verified read-only against the jig checkout's actual merged ADRs before writing:

• chore: add standard repo check surface #1 — sync describe() vs async proof → settled by ADR 0022.
  jig/docs/design/decisions/0022-phase-6-real-driver-integration.md,
  Decision 3: prove-then-describe — the confinement proof runs async at compose time outside
  describe(); describe() stays a synchronous pure getter over an already-computed
  attestation. No P5-pinned seam surface flexes. Matches the brief's expected disposition
  exactly. Confirm holds at the transport-ADR session.
• docs: add org agent contract, roadmap, and Node/pnpm policy #2 — action-union / provenance record impact → settled by ADR 0023 + ADR 0024.
  ADR 0023
  Decision 2 repairs LandingRequest.action from the single mis-encoded literal
  'push|open-pr|merge' to the real union 'push' | 'open-pr' | 'merge' — a local port-type
  fix, freezes nothing, dry-run goldens stay byte-identical.
  ADR 0024
  Decision 3 widens CandidateWorkItem.provenance from the single literal 'jig-validated'
  to an origin-bearing shape, legible in the record as an additive field. Both ADRs confirm
  no observability-records v0 field change was needed — matches the brief's expected
  disposition, citing both ADRs the brief named.
• docs: document the repo-template and apply-repo-standard script #3 — tamper-evidence contract field → resolved for now by ADR 0025.
  ADR 0025
  Decision 1 + Contract Impact Gate Q5: a separate, non-golden integrity sidecar
  (runs/<id>/integrity.json) with an environment-keyed HMAC — no digest/HMAC field added to
  the observability-records v0 contract. Matches the brief's expected "sidecar, no contract
  field" disposition; the ADR explicitly names open question 3 and defers final freeze
  posture to the v0-freeze checkpoint (T14).

All three dispositions matched the brief's stated expectations; no ADR contradicted the
expected reading.

Contract-owner role named (item 8, added on review — G-OWN gate)

The post-spine decisions list also names the contract-owner role: the contract owner for
the two org seams jig owns (execution-plan, observability-records) — the role that receives
routed seam-shape questions (for example, the transport session-observability question and
any port-surface flex) and that alone executes any v0 contract freeze at the T14
checkpoint — is jig's design authority: the org/product owner deciding through jig's
ADR flow (docs/design/decisions/), per M1's seam ownership. Delivery tracks and
sub-agents do not decide seam shapes (the repo-plan "must not decide" boundary).

Gate

pnpm check green in the worktree:

$ pnpm lint
$ pnpm format:check
$ prettier --check "**/*.{md,yml,yaml,json}"
Checking formatting...
All matched files use Prettier code style!

Merge

Do not merge. Do not enable auto-merge. This PR is human-reviewed and human-merged per
decision 5 above — the one wave-1 PR excluded from the autonomous-squash-merge
authorization.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 49d1906c87

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".