Status: draft | Owner: this plan | Created: 2026-01-01
Baseline: main @ 0000000
Proof root: proof/plan-slug/
Next action: see Current resume state
Updated: 2026-01-01 UTC.
- In progress: none. This plan is a draft. [Per task: ID link, owner.]
- Next action: complete the promotion gate.
- Worktree: [per in-progress task: path, branch, HEAD, dirty-file inventory or clean].
- Verification: none. No implementation started.
- Running processes: none.
- Required reads: [in-progress task contracts and relevant current proof].
- Unresolved: [decision or blocker, or none].
One quotable paragraph that states the invariant this plan delivers. A reviewer holds the whole plan against it.
Before:
[component-a: owns policy and transport]
After:
[component-a] -> [policy seam: owns the contract]
- Owns: the concrete workstream this plan delivers.
- Does not own: adjacent work, with the owning plan named for each exclusion.
- Non-goals: work this plan refuses, so scope cannot drift into it.
Promote this plan to active only when every item holds:
- Scope and non-goals are complete.
- Every task states verifiable acceptance criteria.
- The baseline commit and the proof root exist.
- The goal block names real documents, paths, and gates.
- The owner approves activation.
- A numbered condition the implementation cannot trade away.
| ID | Task | Status | Evidence |
|---|---|---|---|
| AB0 | Baseline: pin the baseline commit, create the proof root, author the verifier red, capture fail-before evidence. No production behavior changes. | todo |
|
| AB1 | First implementation task. | todo |
|
| AB9 | Cleanup after the final pull request merges: delete this plan, its proof root, and its index entry. Archive instead only when the repository keeps an archive. | todo |
- Problem: the plan needs a pinned baseline, a red verifier, and fail-before evidence.
- Owning seam and paths: this plan and
proof/plan-slug/. - Steps: pin the baseline commit, create the proof root, author the verifier red, capture fail-before evidence.
- Acceptance: the verifier reports its failing baseline count with no production behavior change.
- Fail-before: the verifier reports red before any implementation task.
- Verification: run the verifier and record its summary line.
- Problem: the defect or gap this task closes.
- Owning seam and paths: module and files.
- Steps: numbered actions.
- Acceptance: named tests or measurable state assertions.
- Fail-before: the check that must fail before the change.
- Verification: exact commands.
- Problem: a merged plan must not linger as a stale control plane.
- Owning seam and paths: this plan file, its proof root, and the plans index.
- Steps: confirm the merge of the final pull request, then delete the plan, the proof root, and the index entry.
- Acceptance: the repository holds no reference to this plan, or the archive holds the completed plan.
- Fail-before: not applicable, because the ledger row records the merge.
- Verification: search the repository for the plan slug and confirm the end state.
Execute this plan to completion. This is a whole-plan goal, not a
single-task goal. Read the compact plan fully, then the in-progress task
contracts and relevant current proof. Required reference owners:
[documents by scope]. Work in [worktree path] on branch [branch]. Chat history
is not progress state. Resume from Current resume state, the ledger, and
verified git state.
Read history only for a named evidence question.
If compaction happens, continue from the plan and git state rather than
restarting. Loop: run independent tasks in parallel. Mark each running task
in_progress with its named owner and its own worktree and branch. A task
starts only when its dependencies are terminal. For each task, implement at
the owning seam, capture fail-before evidence, run the verification commands,
commit the work per the commit policy, write the proof file, append the
execution log with the work commit, mark the task terminal with
evidence, commit the plan update the same way, then start the next eligible
task. Decide rather than ask. Mark a
wrong or already-satisfied task no-action with a one-line reason. Record
a blocker and continue with the next eligible task. Binding constraints:
[invariants and non-goals]. Commit policy: [commit policy]. Stop only at
a valid stop state from the plans skill. Before you stop, update the
ledger and the log, and replace Current resume state. The status line links
to its exact next action. The
goal is met when [completion gate].
Append short rows at the end. This section stays last. History: none. Move older rows unchanged to the proof root when the context budget requires it. Keep a link here.
| Date | Item | Action | Evidence |
|---|---|---|---|
| 2026-01-01 | meta | authored | Plan authored. No implementation started by this step. |