feat(goal): add ordered multi-milestone continuation and cleanup handoff

[BaseInfinity]

Feedback type

Feature request from a successful overnight Codex Desktop Goal-mode run.

Concise description

Add an explicit ordered multi-milestone continuation contract to the generated repo-scoped SDLC workflow. Single-issue completion must not be mistaken for goal completion. After the final milestone, produce a non-destructive branch/worktree residue handoff instead of either leaking clutter silently or deleting user state.

Local context

A private Windows consumer repository requested five documentation milestones in an exact order under one goal. Each milestone required a fresh default-branch read, live issue/milestone read, frozen candidate, TDD evidence, proportional review, PR, hosted CI, merge, issue/milestone closure, and post-merge lifecycle reconciliation. The run succeeded across all five and preserved an unrelated dirty checkout plus a dirty live-work worktree.

No source code, private evidence, paths, secrets, or business data were inspected or included in this report.

Evidence

• Five ordered milestones completed without stopping after the first.
• Five candidate PRs and five lifecycle-only PRs merged.
• The required suite remained green as it grew from 2,214 to 2,275 tests.
• Actual Fable-high review ran where the risk tier required it.
• The exact owner/provider priority remained unchanged throughout.
• The run intentionally retained milestone branches, leaving ten new branch refs whose future retention policy was unclear.

What worked

• A user-supplied exact order plus explicit “do not stop after one issue” was sufficient for Codex Goal mode to continue.
• Re-reading live default-branch and GitHub state before each slice prevented stale execution.
• Per-milestone frozen candidates and separate lifecycle reconciliation kept claims honest.
• Explicit stop conditions prevented provider, tenant, publication, private-evidence, and destructive actions.

Gap

The current SDLC contract is strong for one bounded issue but does not define a reusable multi-issue loop:

1. how the ordered scope is frozen;
2. when one milestone is complete but the overall goal remains active;
3. what state must be re-read before the next milestone;
4. how compaction/resume recovers the exact next item and stop conditions;
5. how candidate and lifecycle receipts compose into overall goal completion;
6. how retained branches/worktrees are reported after completion.

Desired behavior

Add an opt-in ordered_goal contract for GitHub-backed repositories:

• ordered issue/milestone identifiers and an invariant priority owner;
• explicit global stop conditions and per-slice scope;
• before-slice live refresh of default branch, issue, milestone, AGENTS, and ROADMAP;
• candidate → proof → review → PR/CI → merge/closure → lifecycle reconciliation loop;
• overall completion only after every ordered item is terminal;
• compaction/resume carry-forward containing completed items, current item, next item, frozen candidate identity, and stop conditions;
• final read-only inventory of created branches/worktrees with retain, cleanup-candidate, or protected-user-state classification;
• no automatic branch/worktree deletion.

Acceptance criteria

• A disposable fixture completes at least three ordered issues without stopping after issue one.
• A failed/blocked slice preserves completed prior slices and the exact next action.
• Compaction/resume cannot reorder, omit, or repeat a completed milestone.
• Default-branch and live GitHub state are refreshed before every slice.
• Single-issue completion and overall-goal completion are distinct machine-readable states.
• Candidate and lifecycle-only receipts bind to exact trees/PRs.
• Final output reports created branches/worktrees and cleanup eligibility without deleting anything.
• Existing dirty checkouts, untracked files, historical worktrees, and unrelated branches remain untouched.
• Repositories without GitHub continue using the ordinary bounded SDLC flow.

Related

• Teach and enforce an issue-to-PR-to-milestone delivery lifecycle #100 — issue/PR/milestone lifecycle
• feat(lifecycle): make terminal status declarative and reconcile it after merge #136 — declarative terminal reconciliation
• feat(status): add an opt-in repository-wide issue/milestone progress dashboard #135 — progress dashboard
• feat(review): add heartbeat, timeout, cleanup, and structured verdict metadata #137 — structured review metadata
• Audit host-managed memory versus checked-in repository sources of truth #129 — compaction and host-managed memory