plan(seed): Fable 5 long-range remediation roadmap (planning-only, drafts)

.llm/2026-08-08-fable5-remediation-seed-run.md

@@ -0,0 +1,16 @@
+# 2026-08-08 — Fable 5 remediation roadmap seed run (session record)
+
+Run: `.llm/runs/plan-fable5-remediation-roadmap--seed/` · PR #1347 (draft, never merges itself)
+· branch `plan/fable5-remediation-roadmap` · baseline `fac9e339042c` (unchanged at lock).
+
+One-session planning-only seed run (Fable 5 · high supervisor; PLAN-EVAL/IMPL-EVAL owner-waived;
+Claude Workflows with 25 Opus 5 research/drafting subagents across 3 pre-committed workflow
+scripts, 0 errors, ~3.8M subagent tokens). Produced the complete long-range remediation plan
+under `fable-5-remediation-plan/`: 19-artifact cited corpus, synthesis, master plan with a
+12-fork owner sweep, milestone train (two inserted milestones via house rename), 41 issue
+drafts in three milestone directories, RFC-A/RFC-B drafts, 16 amendment blocks, Wave-7 design,
+implementation handoff. Zero GitHub board mutation; owner ratifies filing later.
+
+Lessons worth promoting (candidates, not promoted here): (1) drafting agents that re-verify
+corpus claims at source caught six corrections a prose-only pass would have shipped; (2) the
+"commit workflow scripts before execution" rule made a 25-agent run auditable from the PR alone.

.llm/runs/plan-fable5-remediation-roadmap--seed/context-pack.md

@@ -0,0 +1,32 @@
+# Context Pack — plan-fable5-remediation-roadmap--seed
+
+## Status: COMPLETE (Stage H owner-ratified filing, 2026-08-08).
+
+Planning-only seed run; all charter deliverables landed under
+`.llm/runs/plan-fable5-remediation-roadmap--seed/fable-5-remediation-plan/`:
+
+- `MASTER-PLAN.md` — product bar, principles, target architecture, DAG, sequencing, **12-fork
+  owner sweep (§7)**, risks, exclusions.
+- `ISSUE-DEDUP-AND-SUPERSESSION.md` — dispositions for every relevant existing issue; 41 new
+  drafts tabulated; Stage-E ledger (§3) with six corpus corrections + ownerless items.
+- `MILESTONE-TRAIN.md` — two-milestone insertion via house rename; complete move ledger.
+- `milestones/{0.0.6,0.0.7,0.0.8}-*/` — 41 complete issue drafts (all DRAFT-marked).
+- `rfcs/RFC-A-sdk-client-composition.md`, `rfcs/RFC-B-command-composition-kit.md`.
+- `EXISTING-ISSUE-AMENDMENTS.md` — 16 quotable amendment blocks.
+- `WAVE7-AND-AGENT-ADOPTION.md` — A/B/C measured smoke design; T7-01 exit-gate draft.
+- `IMPLEMENTATION-HANDOFF.md` — filing gate, first PR groups A–D, harness profiles, brief
+  skeleton.
+- `SYNTHESIS.md` + `research/` — 19-artifact cited corpus (waves, board, repo, external).
+
+## Key facts a resuming session must not re-derive
+
+- Draft PR **#1347** is the commit trail; labels moved research→plan at lock. Never merge it.
+- PLAN-EVAL + IMPL-EVAL **owner-waived** (drift D-2); no evaluator was launched; the waiver does
+  NOT extend to future implementation runs.
+- **GitHub filing completed after owner ratification.** The live board is authoritative; see
+  `fable-5-remediation-plan/FILING-LOG.md` for milestone and issue receipts.
+- Baseline `fac9e339042c` verified unchanged at lock. Workflows (committed pre-execution):
+  `wf_e2194004-808`, `wf_03b88126-e7e`, `wf_ebfe8327-306` — 25 Opus 5 subagents total, 0 errors.
+- Six corpus corrections (dedup §3.1) override the raw research files where they conflict.
+- 41 live issues: #1348–#1388. RFC entry points: #1348 and #1361. Implementation sequencing
+  remains forks F1–F12 in `MASTER-PLAN.md` §7.

.llm/runs/plan-fable5-remediation-roadmap--seed/drift.md

@@ -0,0 +1,39 @@
+# Drift Log — plan-fable5-remediation-roadmap--seed
+
+Append-only. Newest last.
+
+## D-1 (2026-08-08, minor, owner directive)
+
+Owner overrides `planning_decisions` effort from lane-policy default **low** to **high** for this
+run (long-range meta-framework roadmap). Recorded in `supervisor.md` § overrides.
+
+## D-2 (2026-08-08, significant, owner directive)
+
+Owner explicitly **waives PLAN-EVAL and IMPL-EVAL** for this planning-only seed run. The seed-run
+stage G hard stop (`plan-eval.md = PASS` before any board mutation) is vacuously safe because the
+run performs **zero board mutation** by charter; the owner personally reviews the plan and decides
+on later adversarial passes and filing. No formal evaluator, OpenHands, OpenRouter, or substitute
+evaluator session is launched. Recorded in `supervisor.md` and `worklog.md`.
+
+## D-3 (2026-08-08, minor, owner directive)
+
+Claude Workflows for research/synthesis fan-out run **Opus 5** subagents (owner directive),
+overriding the `claude_workflow` lane default (Opus 4.8 · low). Contributors only, never
+evaluators; supervisor reviews all output before commit. This also invokes the CLAUDE.md
+documentation-authoring-adjacent exception: the workflow lane touches **no `packages/`/`plugins/`
+source** — output is run-dir planning artifacts only.
+
+## D-4 (2026-08-08, minor, charter deviation from seed-run exemplar)
+
+Stage F (adversarial) and stages G–H are not executed in this run by owner direction. The run ends
+at plan-lock + handoff artifacts (stages A–E + I equivalent). The deliverable set is the
+`fable-5-remediation-plan/` subtree requested in the charter, which supersedes the exemplar's
+folder taxonomy (permitted — seed-run.md fixes stage contracts, not folder names).
+
+## D-5 (2026-08-08, significant, owner ratification)
+
+The owner subsequently ratified the locked plan in-turn and explicitly instructed Stage-H board
+filing. This supersedes D-4 only for Stage H: `FILING-MANIFEST.md` was committed before mutation,
+the milestone train was executed, 41 issues were filed, and existing-owner amendments were added.
+PLAN-EVAL and IMPL-EVAL remain owner-waived for this planning/filing run; the waiver does not carry
+into implementation. See `fable-5-remediation-plan/FILING-LOG.md`.

.llm/runs/plan-fable5-remediation-roadmap--seed/fable-5-remediation-plan/CROSS-RFC-REVIEW.md

@@ -0,0 +1,114 @@
+# Cross-RFC PLAN-EVAL review — RFC-A (#1390) × RFC-B (#1389)
+
+## CYCLE 2 OUTCOME (2026-08-08): **BOTH APPROVED / PASS** — handoff to root for the Qwen pass
+
+Same evaluator session re-evaluated every cycle-1 finding mechanically against the remediated
+texts and worktree source; independent evidence executed (RFC-A fixture `deno check` exit 0;
+section-level anchor verification for all 17 findings).
+
+| RFC | Accepted content SHA | Branch HEAD at verdict | Verdict artifact |
+| --- | --- | --- | --- |
+| RFC-A (#1390) | **`78a7cecd1d5eaafa7a65bc25a21af497567128dc`** | `14b5c858c` (atop author hygiene `9f45404ac` — run artifacts only, verified) | `plan-eval.md` @ `14b5c858c` |
+| RFC-B (#1389) | **`c98c08adabbd992a557ff7c596deae68b9c9cd62`** | `57b51128f` (atop author handoff `6c2043d91`) | `plan-eval.md` @ `57b51128f` |
+
+All F-A1–F-A10 and F-B1–F-B7 resolved (several beyond the asked bar: RFC-A's epoch/reconnect law
+with a preparation-count-2 fixture; RFC-B's per-provider claim-algorithm table). Lane audit:
+`type:test` added to #1390 (compile-only fixture in `packages/sdk/tests/`); `ci:skip-*` ruled
+still valid on both. Both PRs moved `status:plan-eval` → `status:augment-review`.
+
+**Qwen 3.8 Max adversarial acceptance brief (root-owned, separately authorized):** evaluate both
+RFCs at the accepted content SHAs above, unoriented; the two `plan-eval.md` files carry the full
+cycle-1 findings + cycle-2 resolution evidence; the remaining decision surface is the FCP
+question sets (8 in RFC-A, 4 in RFC-B — all adjudicated policy-safe) plus the cross-RFC
+obligations below (§2 amendments, §3 v2-migration epic). Unresolved-by-design: RFC-A FCP Q5
+wrapper-vs-memo, Q6 GET preserve-vs-retire (v2 RFC), metadata Stage 1b ownership; RFC-B FCP
+Q1–Q4. One cosmetic cross-RFC inconsistency remains (frontmatter `target-milestone` convention —
+harmonize at numbering). No board object was created/mutated by this evaluator in either cycle.
+
+The cycle-1 record below is retained for provenance; its "stop" handoff is superseded.
+
+Evaluator: Claude Fable 5 · high, owner-designated cross-family PLAN-EVAL authority
+(2026-08-08). Verdicts of record: `plan-eval.md` in each RFC run dir — **both
+CHANGES_REQUESTED (FAIL_PLAN cycle 1)**: RFC-A @ `f1a29fe1a` (F-A1–F-A10), RFC-B @ `122301d25`
+(F-B1–F-B7). Deep-dive delegations: workflow `wf_b3416478-edf` (3× Opus 5 · xhigh, read-only,
+script committed pre-execution in this run's `workflows/`). This artifact records the
+cross-RFC obligations; it performs **no board mutation**.
+
+## 1. Composition verdict: the two RFCs compose cleanly
+
+- **No circular dependency.** Shared prerequisite #1350 (0.0.7) is one-way: RFC-A stage 1 and
+  RFC-B stage 0 both consume it; neither RFC depends on the other's implementation. RFC-B's
+  service→database edge is new but acyclic (F-B2 requires it be *declared*).
+- **No duplicated error policy.** RFC-A's preparation failures are local, pre-dispatch, never
+  contract errors; RFC-B's command errors are route-opt-in contract errors. Both defer
+  client-visible typing to #1350's literal-preserving spelling. Coherent — with one shared
+  obligation: whichever lands first establishes that spelling; the other must reuse it (F-B7b).
+- **No duplicated context/telemetry policy.** RFC-A reserves trace headers to the transport;
+  RFC-B persists validated W3C context in rows and uses producer/consumer spans — disjoint
+  layers, consistent with the existing telemetry vocabulary. Both correctly avoid inventing a
+  second correlation scheme; RFC-B's stricter-than-existing attribute redaction (F-B7h) is a
+  vocabulary-cleanup question, not a conflict.
+- **Shared `(family, major)` protocol vocabulary** (RFC-A ↔ #928) does not leak into RFC-B.
+- **One latent intersection to watch at implementation:** RFC-A's `idempotency-key` header
+  allowance and RFC-B's envelope `idempotencyKey` are different layers (transport header vs
+  command input). No conflict today; the future HTTP-idempotency recipe should name which one is
+  authoritative for services that accept both.
+
+## 2. Sequencing vs the filed board (#1348–#1388, PR #1347)
+
+Coherent as filed: RFC trackers #1348/#1361 in 0.0.6 (ratification); RFC-A implementation
+children #1349–#1353 in 0.0.7; RFC-B children #1362–#1364 (+#1363 umbrella) in 0.0.8 with #1350
+(0.0.7) as stage 0. The milestone descriptions carry the authority banner. Frontmatter
+`target-milestone` semantics need one clarifying line in each RFC (RFC-A says 0.0.7 = impl,
+RFC-B says 0.0.6 = ratification — pick one convention; F-B7d).
+
+**Existing children are sufficient — no duplicates needed.** Required amendments (owner-ratified,
+not executed by this evaluator):
+
+- **#1351** — add: `@orpc/opentelemetry` rename decision (already available on v1, F-A8d);
+  exact-pin vs lock-only-pin policy for the family move (F-A8, caret manifests); the GET-dedupe
+  no-op trap as an acceptance row (F-A8c).
+- **#1349** — add: prepared-header channel statement, private-port location + `deno doc`-absence
+  gate, server key-algebra surfaces (F-A2/A6/A7) once the amended RFC lands.
+- **#1350** — stage-0 body reconciliation (metadata initialization + literal-preserving spelling
+  shared by both RFCs) — already planned by RFC-A stage 0; keep.
+- **#1362/#1363** — inherit F-B2's relay-ownership decision and F-B1's claim algorithm once
+  amended; #1363's child table gains the `PrismaTransactionClient` generator deliverable (F-B6).
+- **New reconciliation candidate (owner decision):** `@netscript/queue`'s runtime `ensureSchema`
+  DDL vs RFC-B's no-hidden-migrations doctrine (F-B3) — file only if the owner adopts the rule
+  repo-wide rather than kit-scoped.
+
+## 3. oRPC v2 migration: separate RFC epic — recommended, with exact scope
+
+**Recommendation: yes, one new RFC tracking issue (rfc-form, 0.0.6-adjacent ratification,
+implementation unscheduled)** after RFC-A acceptance — not filed by this evaluator. Verified
+facts anchoring it: v1.15.0 is latest stable (shipped *after* beta.26 — v1 actively maintained);
+v2 wire protocol incompatible; 74 non-test files reference `@orpc/*`.
+
+Scope (RFC-A's gate list + the four evaluator additions): keep-or-drop **GET** decision
+(`allowMethods` + Sec-Fetch-Mode CSRF story — direction corrected per F-A8a); re-implement
+`inferRPCMethodFromContractRouter` (removed in v2); dedupe-effectiveness gate (GET-only filter);
+`defineMeta` migration for `NetScriptProcedureMeta`; `errorStatusMap` split vs #1350 spelling;
+middleware-dedup removal audit across the 74 files; OTel span-topology/double-span proof
+(package rename excluded — that is #1351, v1); serializer/streaming/Fresh/desktop parity;
+TanStack key re-verification; coordinated-rollout vs parallel-endpoints owner decision (RFC-A
+Q9); exact-family pinning discipline; full conformance-suite re-run on the v2 adapter.
+Dependencies: RFC-A accepted + stages 2–3 landed (the ports are the migration boundary); owner
+beta-risk decision; #1351 complete.
+
+## 4. Handoff state (root orchestrator)
+
+- **Both verdicts request changes → per the evaluator contract, this session stops here.** The
+  Qwen 3.8 Max adversarial pass waits until both RFCs reach APPROVED.
+- Resume the **RFC-A Codex generator** (thread `019fe242-2bd9-7ff3-8044-bd9d09585397`) with PR
+  #1390's finding list (F-A1–F-A8 required; F-A9/A10 may ride along). Branch
+  `docs/rfc-sdk-client-contribution`, HEAD `f1a29fe1a` (evaluator commit atop `7be129d80`).
+- Resume the **RFC-B Codex generator** (thread `019fe242-2c45-7e03-a428-eebfb968eda0`) with PR
+  #1389's list (F-B1–F-B4 required; F-B5–F-B7 as batch edits). Branch
+  `docs/rfc-command-composition-kit`, HEAD `122301d25` (evaluator commit atop `62304176f`).
+- Both PRs moved to `status:plan`; restore `status:plan-eval` with the amended handoffs. Cycle 2
+  is the last before escalation (two-FAIL_PLAN limit).
+- Evaluator wrote only: the two `plan-eval.md` files (pushed with explicit refspecs to their own
+  branches), the two PR comments + label moves, this artifact, and the seed-run worklog note. No
+  RFC text, no product code, no issues/epics/milestones touched. RFC-B's pre-existing dirty
+  `codex-thread-ids.md` was left untouched.