docs(roadmap): correct #347 — native /goal exists; add #350 cadence-gap fix

ROADMAP.md

@@ -296,7 +296,8 @@ Living tracker of projects shipped using this wizard. **Rule:** only list projec
 | 235 | ~~Research: Thoughtworks "AI Evals" methodology vs our existing E2E pipeline~~ ✅ DONE 2026-05-04 — see [`.reviews/research-235-ai-evals.md`](.reviews/research-235-ai-evals.md) | **Verdict: NO-GO.** Audit confirms the article is methodology-only (no tools/frameworks named); every layer Thoughtworks describes (pre-deployment validation, post-deployment monitoring, quality gates, continuous oversight, performance consistency, output accuracy, error-mode catching, "dynamic nature of AI") already has a working analog in our pipeline (Tier 1/2 evaluator + score-history + CUSUM + token-spike-check + adversarial cross-model review + SDP scoring). The only candidate gap is bias/alignment evaluation, which is explicitly out of scope (the wizard is SDLC enforcement, not LLM ethics — that belongs to model providers and domain-specific tools). Pattern continues with #76 (Promptfoo NO-GO) and #77 (constrain-to-playbook NO-GO): external-methodology audits keep validating our pipeline is mature; revisit only when an external source points to a *specific* technique we don't have. **Original question:** does Thoughtworks' AI Evals framework ([thoughtworks.com/en-us/insights/decoder/a/ai-evals](https://www.thoughtworks.com/en-us/insights/decoder/a/ai-evals)) add anything we don't already have in our CI/CD? User call-out 2026-05-01: "should we be using AI Evals like Thoughtworks defines? if so add to roadmap or research." |
 | 236 | ~~Audit: functional bloat hunt across wizard surface~~ ✅ **PHASE-PARTIAL DONE 2026-05-05; +tracked-artifact sweep 2026-05-06.** Three token-amplifier fixes shipped (v1.69.0 BASELINE-once-per-session ~12K saved/heavy-session, v1.70.0 TDD CHECK-once-per-session ~0.5-1.5K saved, v1.71.0 SDLC SKILL.md Cross-Model Review trim 4995 → 4568 tokens) plus one **orphan deletion** (`scripts/audit-subagent-models.sh`, 85 lines, 0 references, never in npm `files` list). **v1.73.0 (PR #331)** added category-(d)-adjacent sweep: -460 LOC across 15 stale `.reviews/` round-N + preflight artifacts plus `plans/CATCHUP.md` — all gitignore-shadowed leftovers from pre-`.gitignore` commits, none load-bearing. Live evidence in maintainer session: BASELINE block disappeared from system-reminders mid-conversation when sentinel landed. **Categories still not exercised:** (b) hooks (no nudge proven noise), (c) tests redundancy pass, (f) ROADMAP archive-below-fold. Re-trigger when a maintainer says "this repo is hard to find things in" again. **Original scope below for re-runs:** Recurring "does this still earn its keep?" pass. **Distinct from neighbors:** #220 token-bloat is size-per-file (chars/tokens); #19 Monthly GC is "weak tests / test fluff." This is **per-component existence justification**. **Method:** Prove-It style — for each candidate, attempt `git rm` on a branch, run full test suite. If nothing fails, that's a real bloat candidate. **Trigger:** quarterly. User call-out 2026-05-04: "bloat sucks." |
 | 302 | User-level setup-wizard + repo-local lifecycle split — **DESIGN-CERTIFIED, implementation deferred** | GH issue #302 proposes splitting the wizard so `setup` lives user-level/global and `sdlc`/`update`/`feedback` stay repo-local ("install once, bootstrap anywhere"). Cross-model design review completed 2026-05-20 (Codex gpt-5.5 xhigh, [`.reviews/302-design-codex-review.md`](.reviews/302-design-codex-review.md)) scored Claude's first-pass analysis 5/10 NOT CERTIFIED and replaced it with a concrete channel contract. **Decided channel split:** plugin channel = user-level/global path (already exists: `.claude-plugin/plugin.json` + `hooks/hooks.json` with `${CLAUDE_PLUGIN_ROOT}`); npm/npx = explicit repo-local bootstrap/update path; no npm `postinstall` writing to `~/.claude/` (fragile under CI/containers/read-only HOME/Windows). **Design questions answered:** (1) `setup` global via plugin, not also copied repo-local (precedence #338 makes the global copy dead weight when both exist); preserve auditability via `SDLC.md` metadata or install manifest, not duplicate same-name skill. (2) `feedback` global via plugin, can still read repo-local `SDLC.md` at invocation time. (3) `update-wizard` stays repo-local (reads `SDLC.md` metadata, runs drift detection, mutates project files). (4) `npm install -g` installs the CLI binary only — no lifecycle scripts. (5) Migration: leave existing repos unchanged; new split is opt-in only; later major release can stop copying `setup`/`feedback` repo-local for new installs. **Hard constraint Claude missed:** "global setup auto-triggers anywhere" is FALSE — `hooks/instructions-loaded-check.sh:14-22` and `hooks/sdlc-prompt-check.sh:27-35` exit silently when `SDLC.md`/`TESTING.md` don't exist, and `tests/test-hooks.sh:2507/2522` actively assert that silence. A plugin-only fresh-repo bootstrap signal must be designed before the global skill can satisfy the "always available" promise from the issue. **Existing dual-channel surface to respect:** `cli/init.js:286-302` blocks npm-on-plugin (duplicate `/update-wizard`), `hooks/instructions-loaded-check.sh:195-206` warns on dual install. Adding npm-global skills would create a third overlapping surface — deprecate one channel first. **Implementation sequence (5 steps, do NOT start without dedicated session):** (1) lock the channel contract in `CLAUDE_CODE_SDLC_WIZARD.md` + ARCHITECTURE.md; (2) doc the fresh-repo trigger limitation + migration story in #302 comment; (3) prototype plugin-only fresh-repo bootstrap signal + assert current silent-hook tests still pass; (4) write global-skill collision/backup/update tests if any installer touches `~/.claude/skills/`; (5) only then alter `init` behavior for new installs, leaving existing ones intact. **Why deferred 2026-05-20:** session shipped v1.74.0 → v1.75.0 → v1.75.1 (Trusted Publishing migration), maintainer fatigued, #302 is architectural — needs a fresh planning session. **Trigger to revisit:** explicit user request OR a second repo install where "install once" friction is felt acutely. |
-| 347 | Goal-mode checkpoint workflow (Codex `$gdlc` equivalent) — **RESEARCH-DONE, implementation deferred** | GH issue #347 asks whether Claude Code has a native primitive for long-running goal-bound work (Codex `$sdlc + $gdlc` pattern with persistent constraints, checkpoint cadence, explicit stop boundaries). Research completed 2026-05-23 ([`.reviews/347-goal-mode-research.md`](.reviews/347-goal-mode-research.md)) via claude-code-guide subagent against CC docs current as of 2026-05-23. **Verdict: NO native primitive.** CC has no first-class abstraction for a long-running multi-session objective with enforced lifecycle constraints. Sessions are independent JSONL files; persistence is indirect via `CLAUDE.md`, auto memory, hooks. **Per-feature mapping:** TaskCreate = per-session only; subagents = per-invocation context only; skills = standing instruction for current session only; hooks = YES the deterministic enforcement layer; plan mode = per-turn only; `CLAUDE.md` = context not enforced configuration; no `/goal` command on roadmap. **Closest existing primitive:** user-invocable skill + `SessionStart` hook + project `GOAL.md` — exactly what the wizard already does for SDLC enforcement, just unwrapped for goal-mode semantics. **Prove-It Gate verdict:** absorb into existing `/sdlc` skill as a "Long-Running Goal Mode" section + `GOAL.md`/`HANDOFF.md` templates rather than building a new `/goal-start`/`/goal-handoff`/`/goal-stop` skill trio. Default to absorption per #236 functional bloat principle; promote to dedicated skill only if absorbed form proves too easy to ignore. **Implementation sequence (5 steps, do NOT start without dedicated session):** (1) add `GOAL.md` + `HANDOFF.md` templates to `cli/templates/` and prompt for them in setup-wizard's long-running-project branch; (2) extend `/sdlc` skill with "Long-Running Goal Mode" section (objective binding, checkpoint cadence, stop-boundary enforcement); (3) extend `hooks/instructions-loaded-check.sh` to inject `GOAL.md` contents via `additionalContext` when present; (4) extend `hooks/sdlc-prompt-check.sh` to re-assert stop boundary if `GOAL.md` declares one and context is >75% full; (5) quality tests proving goal injection fires only when `GOAL.md` present, stop-boundary reassertion fires only past threshold, `HANDOFF.md` write is idempotent across resume sessions. **Why deferred 2026-05-23:** only one demand signal (issue author); Prove-It Gate "Default answer is NO unless proven YES" applies. Per #233 pattern, build when a second user files for it OR when the maintainer's own work hits the goal-mode gap. **Trigger to revisit:** second user request, OR maintainer files a session-spanning project where the absence of `GOAL.md` is visibly painful. |
+| 347 | Goal-mode checkpoint workflow (Codex `$gdlc` equivalent) — **CORRECTED 2026-05-24: native `/goal` exists, scope shrunk to ~30-line skill wrapper** | GH issue #347 asks whether Claude Code has a native primitive for long-running goal-bound work (Codex `$sdlc + $gdlc` pattern with persistent constraints, checkpoint cadence, explicit stop boundaries). **2026-05-23 research was WRONG** — claude-code-guide subagent said no native primitive, but CC **v2.1.139 shipped native `/goal`** (confirmed via raw changelog curl; docs at [code.claude.com/docs/en/goal.md](https://code.claude.com/docs/en/goal.md); follow-up fixes v2.1.140 hook-disabled hang + v2.1.143 subagent race). Corrected research at [`.reviews/347-goal-mode-research-CORRECTED.md`](.reviews/347-goal-mode-research-CORRECTED.md). **What `/goal` does:** session-scoped, evaluator-driven (Haiku default judges transcript after each turn yes/no), survives `--resume` but not `/clear`, no disk writes, no native turn/time cap. UX = `/goal <condition>` + `/goal` (status) + `/goal clear`. **Old verdict OBE.** **Corrected scope = ~30-line skill wrapper in existing `/sdlc` (NOT a `GOAL.md` template — `/goal` writes nothing to disk).** Per [`.reviews/347-goal-mode-research-CORRECTED.md`](.reviews/347-goal-mode-research-CORRECTED.md), the wizard should add: (a) **pre-flight checklist** — workspace trusted, hooks not disabled at any settings layer, CC ≥ v2.1.143; (b) **condition-writing guidance** mirroring SDLC reporting standards (measurable end state + check + constraints + hard turn/time bound since `/goal` has no native cap); (c) **compose-with-hooks note** — `UserPromptSubmit`/`SessionStart`/`PreCompact` fire normally inside the goal loop, so `sdlc-prompt-check.sh` + `precompact-seam-check.sh` keep gating each turn; (d) **resume caveat** — `--resume` resets turn/time counters; (e) **anti-pattern callout** — don't use `/goal` for "doneness" the evaluator can't see in the transcript (evaluator can't call tools). **Status:** Actionable now (qualifies as "low-cost cleanup" under demand-signal-first gate — ~30 lines, zero-API, builds on existing skill, corrects a wrong public answer). **Original 5-step plan + `GOAL.md` template scaffolding OBE.** **Meta-lesson captured in CORRECTED doc:** require explicit citation of an authoritative source (docs index, raw changelog grep) before accepting a "feature does not exist" claim — negative claims are easier to fake than positive ones. **Process gap also caught:** auto-update PR workflow stopped flagging features at CC v2.1.118 (2026-04-23, last PR #210) because #231 Phase 3d (v1.54.0) gutted the in-CI LLM-ranker to $0. Manual replacement (`claude --print --allowedTools "WebFetch,Read,Bash" "$(cat .github/prompts/analyze-release.md)"`) was supposed to run weekly on Max — it didn't, which is how `/goal` slipped past unnoticed for ~5 weeks. See #350 for the cadence fix. |
+| 350 | CC feature-discovery cadence — replacement-for-#231 ranker has no enforced cadence | **Process gap discovered 2026-05-24 via #347 correction.** #231 Phase 3d (v1.54.0, 2026-04-29) deleted the in-CI Claude ranker from `weekly-update.yml` to take it to $0/week. The fallback was "maintainer runs `claude --print "$(cat .github/prompts/analyze-release.md)"` weekly on Max." That ran zero times in 5 weeks. Consequence: native CC `/goal` shipped in v2.1.139 (~mid-May) and was missed until a user asked a related question 2026-05-24 — wizard then published wrong research saying "no /goal exists." 39 CC versions accumulated since the last auto-PR (v2.1.118 → v2.1.150). **Scope (Actionable now, low-cost cleanup):** (a) restore a thin cron — `weekly-update.yml` still runs detection ($0 GH API call); add a single follow-up step that opens a "CC version drift" GitHub issue when our `<!-- SDLC Wizard Version -->` baseline is >5 minor versions behind latest npm — pure GitHub API, no LLM, no API spend; (b) the issue body links the maintainer to the on-Max command + the changelog URL + a one-line "what to do" reminder; (c) session-start `instructions-loaded-check.sh` already has the staleness nudge (#196 — "N minor versions behind") — verify it actually fires when the gap is ≥3 minor AND link to the analyze-release runbook in the loud branch. **Why this beats #85 Phase 2** (which we killed 2026-05-24): #85 Phase 2 was "auto-rank features in CI via LLM" — expensive and conflicted with #231. This is "open a stale-baseline issue when we detect drift" — cheap, GH API only, just a maintainer reminder. **Status:** Actionable now. **Trigger to do:** immediately, paired with #347 since both came from the same discovery. |
 
 ## Review Pipeline