feat: harvest 3 P4 — the github-issues axis, after the draft, decision records, hygiene, what the phases emit (C5–C7, C9, C10)

.claude/agents/adr-conformance-reviewer.md

@@ -27,7 +27,7 @@ If only a branch name is given, run `git diff main...HEAD` (or against the named
    - **A boundary impl** (e.g., a new backend, driver, or adapter) → the ADR that defines the behaviour/interface
    - **A new top-level module or service** → architectural ADRs about layering, hot-path vs cold-path placement, or extension seams
 
-3. **Read each intersecting ADR — and follow its `Refines:` / `Supersedes:` chain.** Don't skim; the "Decision" and "Consequences" sections are the contract. If an intersecting ADR is refined or partially superseded by a later one (named in the index, or in a child's `Refines:` / `Supersedes:` field), **load the child too**. A *refiner* narrows a parent clause to a scoped reading *without* superseding it (the parent stays Accepted); a *partial supersession* replaces one named clause while the rest of the parent stands. A parent's literal clause is often narrower in practice than it reads — the scoped reading in the chain is authoritative.
+3. **Read each intersecting ADR — and follow its `Refines:` / `Extends:` / `Supersedes:` chain.** Don't skim; the "Decision" and "Consequences" sections are the contract. If an intersecting ADR is refined or partially superseded by a later one (named in the index, or in a child's `Refines:` / `Supersedes:` field), **load the child too**. A *refiner* narrows a parent clause to a scoped reading *without* superseding it (the parent stays Accepted); a *partial supersession* replaces one named clause while the rest of the parent stands. An *extender* adds an obligation beside a parent clause that stays satisfied — the asymmetric case: the parent passes unchanged while the child can fail, so a diff clean against the parent is not clean until every extender is checked. A parent's literal clause is often narrower in practice than it reads — the scoped reading in the chain is authoritative.
 
    **Also read the project's reconciliation layer, if one exists.** Some projects keep a buildable reconciliation document that turns frozen upstream references + charter ADRs into the as-built spec; a scoped reading there is authoritative for the build. Read it before flagging an apparent ADR-literal violation.
 
@@ -52,6 +52,7 @@ If only a branch name is given, run `git diff main...HEAD` (or against the named
 - **Be specific.** "Violates ADR-NNNN" is useless without "because the LLM is acting as a controller (line 42 calls `Servo.set_angle/2` directly) instead of emitting a tool call".
 - **Don't flag what isn't a violation.** ADRs don't govern every line; if a change is orthogonal, it's orthogonal. False positives erode trust in this reviewer.
 - **A literal ADR clause may be refined — confirm before flagging.** Before reporting "violates ADR-NNNN Dn", check that no `Refines:` child ADR or reconciliation-layer entry narrows that clause for the case at hand. A flag against a correctly-scoped clause is a false-positive, and false-positives erode trust faster than misses.
+- **Consult `docs/adr/corrections.md` before flagging a claim.** A citation, figure, attribution or formula already recorded wrong there is cited by its `C-` number, never restated as a finding; a wrong claim not yet recorded is reported as one, with the register named as its destination.
 - **Don't propose new ADRs.** That's an `/adr-new` invocation, not a review finding.
 - **Don't review style, tests, or logging.** Other reviewers cover those. Stay scoped to ADR conformance.
 

.claude/agents/cascade-rule-reviewer.md

@@ -15,7 +15,7 @@ The `adr-conformance-reviewer` covers ADRs in `docs/adr/`. The `logging-discipli
 
 The contracts I check for compliance:
 
-Every file under `.claude/rules/` **except** `logging.md` (the logging reviewer's) and the `pr-review.md` pair (the rubric I apply, not a contract I check the diff against). That set is derived, not enumerated — a rule added, split or renamed is in scope the moment it lands. Today it is: `testing.md` (three-regime testing; the named tests trace to acceptance criteria), `cbk-conventions.md` and `cbk-conventions-reference.md` (branch naming, title prefixes, close markers, the `[skip ci]` rule, mutation discipline, workstream slug stability), `simplification.md`, `knowledge-backend.md` (writes HITL-gated and default-SKIP; the repo is canonical for code and cascade artifacts), `workflows.md`, `tooling.md`, `orchestration.md` and `orchestration-reference.md`. Read the reference half of a split rule when the contract's pointer names the section the diff touches.
+Every file under `.claude/rules/` **except** `logging.md` (the logging reviewer's) and the `pr-review.md` pair (the rubric I apply, not a contract I check the diff against). That set is derived, not enumerated — a rule added, split or renamed is in scope the moment it lands. Today it is: `testing.md` (three-regime testing; the named tests trace to acceptance criteria), `cbk-conventions.md` and `cbk-conventions-reference.md` (branch naming including the issue-less form, title prefixes, close markers, the `[skip ci]` rule, mutation discipline, workstream slug stability, and the reference half's § Licensing, § Dependency settle-window and § .gitignore anchoring), `simplification.md`, `knowledge-backend.md` (writes HITL-gated and default-SKIP; the repo is canonical for code and cascade artifacts), `workflows.md`, `tooling.md`, `orchestration.md` and `orchestration-reference.md`. Read the reference half of a split rule when the contract's pointer names the section the diff touches.
 
 Out of scope (handled by other reviewers):
 - `.claude/rules/logging.md` → `logging-discipline-reviewer`

.claude/commands/enrich.md

@@ -41,7 +41,7 @@ Synthesize into the resolved intent + the resolved forks. Don't move to Step 3 u
 
 A read-only investigation, like `/finish`'s plan-mode research — **do not write code here.** Ground the resolved forks in the actual code: the files / symbols the issue touches, the constraints (ADRs, `.claude/rules/*`, `docs/STANDARDS.md`), the dependencies, and the gotchas.
 
-- `Grep` / `Glob` for the surface; find-references / symbol-info / call-site tooling for the blast radius; documentation-lookup tooling to confirm a library's actual behavior; read-only data/schema inspection when the issue is data-shaped.
+- `Grep` / `Glob` for the surface; find-references / symbol info / call sites for the blast radius (which tool: `tooling.md` § Code intelligence); documentation-lookup tooling to confirm a library's actual behavior; read-only data/schema inspection when the issue is data-shaped.
 
 **Scale the investigation to the surface — when the tooling is available:**
 

.claude/commands/finish-procedure.md

@@ -183,6 +183,8 @@ The principle for Steps 8–9: **the executor handles findings the codebase need
      - **Defer (N)** — `<verbatim rationale> — conflicts with <ADR / scope / framing>`.
      - **Reject (N)** — one-line dismissal.
 
+   **The roadmap row, before `gh pr create`** — where the project keeps `docs/cbk/ROADMAP.md` and the spec names the flip: update the row and `## Now` on this branch as the last commit (`docs(cascade):`, no CI-skip marker — it is HEAD at the flip), with the PR number predicted from the shared issue-and-PR sequence (GitHub's REST API considers every pull request an issue and its issue endpoints return both — `https://docs.github.com/en/rest/issues/issues`, read 2026-09-06 — so the newest number across both is the last item of an issues listing) (`gh api 'repos/{owner}/{repo}/issues?state=all&sort=created&direction=desc&per_page=1'` plus one) and checked against the number `gh pr create` returns; one correcting commit on a mismatch; never a separate PR (the contract, item 8).
+
    If `gh pr create --draft` fails (auth, missing scope, draft PRs disabled, rate limit, PR already exists for this branch), stop and surface the failure with per-step status.
 
 ## Step 11: Hand off
@@ -199,8 +201,12 @@ If any finding was classified **Defer** because it would conflict with an ADR or
 
 **Roll-forward offer** (HITL). If the run surfaced deferred context that a *later* unit of work will need — TODOs, contract facts the next issue depends on, intentional omissions, or a follow-up worth its own issue — **offer** to post a roll-forward comment **wherever it's relevant**: the next issue, a sibling, a downstream milestone issue, a meta-issue, or a brand-new follow-up. Do **not** hardwire it to "the next issue." Default to posting **post-merge** unless the deferred context blocks another issue's start. If nothing was deferred, skip this silently.
 
+**Backward sweep** (HITL offer, beside the roll-forward). What this run landed can make a sibling spec's *absence* claim false — "no adapter exists yet", "the table is not wired". The candidate set is closed: the open siblings under the same parent issue carrying `cascade-depth:roughed-in`, read in **one** call (`gh api repos/{owner}/{repo}/issues/{parent}/sub_issues` on github-issues, the parent's children on linear) and capped by the milestone's R-count — never the whole backlog. Check absence claims only, and verify each by **running the body's own command** (its grep, its test, its probe), not by recalling what this run did. Separate the two kinds: a claim this landing falsified (report it, with the commit) from a claim that was false when written (report it as a spec defect). **Default to not reporting** — a claim that still holds is not mentioned. The repair is rough-in's in-place body refresh, offered to the operator; this run never edits an issue body. An empty sweep is a one-line pass — `backward sweep: N siblings, no absence claim falsified` — never an omitted line.
+
 **Optional learning-runbook write to Notion** (only when knowledge backend = `notion` is configured for this repo, per `.cascade/backends.toml`): if this execution surfaced a learning that would be a useful durable cross-project runbook entry (e.g., a non-obvious gotcha with the stack, a recipe for a recurring task spanning repos, a corrected understanding of an external system's behavior), prompt the operator: *"Want to promote this learning to a Notion runbook page under the Engineering Wiki? Defaults to SKIP."* Defaults to **SKIP**. Per `.claude/rules/knowledge-backend.md` § "When to write" — this is the rare opt-in path; HITL-gated; never default. If the operator opts in, announce the planned write (title, parent, body preview) before committing. This step is gated to genuinely cross-project learnings only — local-to-the-PR observations stay in the PR body and the hand-off summary, not in Notion.
 
+**Post-merge checklist** (the hand-off ends with it; the operator runs it after the merge): `git switch main && git pull --ff-only && git fetch --prune`; post the drafted roll-forwards; close the meta-issues the close marker did not; append the frame's `## Rough-in events` row when this PR was the milestone's capstone; flip the roadmap row if Step 10 did not; the next runnable command, read off the roadmap. A research milestone's capstone is the verdict artifact — name it and its owner.
+
 Do not call `gh pr ready` and do not merge — those remain the user's calls. The user's draft-review pass is where Surface findings get decided; they can ask you to apply any of them in the PR-feedback-loop turn, or wave them through.
 
 ## What `/finish` does NOT do
@@ -209,7 +215,7 @@ Do not call `gh pr ready` and do not merge — those remain the user's calls. Th
 - **Does not handle re-rough-in.** Surface the need and return to chat for the rough-in skill.
 - **Does not auto-create dependent issues.** If the work reveals a missing sub-sub-issue, surface and return to chat for rough-in.
 - **Does not bypass dependencies.** Refuse to proceed when dependencies are unmet.
-- **Does not modify cascade artifacts** (`docs/cbk/blueprint.md`, `docs/cbk/frame-NN.md`, etc.). Those are produced by chat-skill cascade phases.
+- **Does not modify cascade artifacts** (`docs/cbk/blueprint.md`, `docs/cbk/frame-NN.md`, etc.). Those are produced by chat-skill cascade phases. *The status surfaces are carved out: `docs/cbk/ROADMAP.md`'s row and a frame's `## Rough-in events` table are not cascade events; Step 10 flips the row on this branch when the spec names it, the post-merge checklist otherwise (the contract wins where the two differ).*
 - **Does not modify ADRs.** ADRs are immutable once accepted; revisions happen via new ADRs that supersede the old (chat-skill territory, not `/finish`).
 - **Does not make workstream-level or framing-level decisions.** Those belong to upstream cascade phases (blueprint, framing, rough-in).
 - **Does not skip `/simplify` or `pr-review-toolkit:review-pr`.** The simplify and review passes both run before the PR opens; the simplify pass is non-negotiable per `docs/STANDARDS.md` § Step 4, and the review-toolkit findings are auto-triaged with the four-class rubric (Apply for correctness/validity/defensive; Surface for taste/style; Defer/Reject for conflicts and agent errors). The one exception is an explicit break-glass marker (Step 1), which waives exactly the `pr-review-toolkit:review-pr` half and is recorded on that skill's `## Review gate` line. The orchestrated sweep supplements the two skills and never substitutes for either.