Add /issues skill and outstanding-work memory ledger

.claude/skills/issues/SKILL.md

@@ -0,0 +1,72 @@
+---
+name: issues
+description: Track and recall all outstanding tasks, recommendations, and issues for this repo as durable cross-session memory. Use when the user types "/issues" (state the open items back), or asks to add/close/update/capture an outstanding task, recommendation, or issue. The memory lives in docs/outstanding-issues.md; a plain "/issues" is read-only.
+---
+
+# issues — the outstanding-work memory
+
+`docs/outstanding-issues.md` is the durable, cross-session memory of everything still outstanding:
+open **tasks**, **recommendations** not yet acted on, and **issues** not yet resolved. Chat context
+resets; that file does not. This skill reads it back and keeps it current.
+
+**The ledger is the source of truth, not chat memory.** Never answer `/issues` from conversation
+recall — always read the file first, so the answer is correct even in a fresh session.
+
+## Trigger
+
+- User types `/issues` (optionally with a subcommand or filter below).
+- User asks to add / close / update / list / capture an outstanding task, recommendation, or issue.
+
+## Default: `/issues` (read-only)
+
+1. Read `docs/outstanding-issues.md`.
+2. State the **open items** back, grouped by priority (P1 → P3), each as
+   `#ID · type · summary — next action (source)`.
+3. End with a one-line count, e.g. `5 open: 0×P1, 3×P2, 2×P3 · 0 resolved this session`.
+4. Do **not** mutate the file or commit on a plain read.
+
+If a filter is given, narrow step 2: `/issues P1` (by priority), `/issues issues` / `/issues recs`
+/ `/issues tasks` (by type), `/issues <keyword>` (summary/detail substring match).
+
+## Mutating subcommands
+
+Parse the intent from natural language too — the exact syntax is a convenience, not a requirement.
+
+- **`/issues add <text>`** — append a row to **Open items**. Infer `Pri`/`Type` from the text
+  (ask only if genuinely ambiguous; default `P2`/`task`). Allocate the ID from the
+  `<!-- issues:next-id=NNN -->` marker, then bump that marker. Fill `Source` with
+  `session <today>` unless the user names one; `Added` is today's date.
+- **`/issues done <id> [outcome]`** — move that row from **Open items** to **Resolved / archive**
+  with today's date and a one-line outcome. Archive, never delete.
+- **`/issues update <id> <text>`** — edit an open row's summary or next action in place.
+- **`/issues capture`** — scan the current session for recommendations, follow-ups, deferrals, and
+  unfixed problems that surfaced but were not recorded. Propose them as a numbered list and add the
+  confirmed ones (dedupe against existing rows first — do not re-add something already tracked).
+
+## Capture discipline (proactive memory)
+
+When a task in _any_ session ends with unresolved follow-ups — a deferred fix, a "revisit when X"
+recommendation, a known risk, a TODO you had to leave — offer to record them here before the context
+is lost. That is what makes this a memory rather than a static list. Prefer one crisp row over a
+paragraph; put the smallest next action in **Detail / next action**.
+
+## Writing rules
+
+- Keep the table format and column order exactly as in `docs/outstanding-issues.md`. One row per item.
+- IDs are monotonic and never reused — always allocate from the `issues:next-id` marker and bump it.
+- Escape `|` inside cell text (write `\|`) so the markdown table stays intact.
+- Respect the repo's RAG/clinical/privacy flagging rules if an item _itself_ touches a protected
+  surface — recording it here is fine, but acting on it later still needs the usual gate.
+
+## Persist the memory (commit)
+
+After any mutation, stage and commit **only** `docs/outstanding-issues.md` so the memory survives the
+ephemeral container and other worktrees:
+
+```
+git add docs/outstanding-issues.md
+git commit -m "issues: <what changed>"
+```
+
+Do not stage or commit anything else, and do not push unless the user asks (or you are already in a
+handoff/upload flow). A plain read-only `/issues` commits nothing.

AGENTS.md

@@ -411,6 +411,20 @@ Run the matching planner command in `docs/productivity-workflows.md` without sid
 
 <!-- END:repo-productivity-skills -->
 
+## Outstanding-work memory (`/issues`)
+
+`docs/outstanding-issues.md` is the durable, cross-session memory of every outstanding **task**,
+**recommendation**, and **issue** for this repo. Chat context resets between sessions; that file does
+not, so anything worth remembering after a session ends belongs there.
+
+- When the user types `/issues`, invoke the `issues` skill (`.claude/skills/issues/SKILL.md`): read
+  `docs/outstanding-issues.md` and state the open items back, grouped by priority. A plain `/issues`
+  is read-only — it mutates and commits nothing.
+- `/issues add|done|update|capture …` mutate the ledger; each mutation commits **only**
+  `docs/outstanding-issues.md` (no push unless the user asks or you are already handing off).
+- Proactively offer to `capture` unresolved follow-ups, deferrals, and known risks into the ledger
+  before a session's context is lost — that is what keeps it a memory rather than a stale list.
+
 ## Codex GitHub review behavior
 
 These instructions apply to Codex GitHub pull request reviews and Codex tasks started from PR comments.

docs/outstanding-issues.md

@@ -0,0 +1,48 @@
+# Outstanding Issues, Recommendations & Tasks
+
+Durable, cross-session memory of everything still outstanding for this repo: open **tasks**,
+**recommendations** not yet acted on, and **issues** not yet resolved. Chat context is ephemeral
+(sessions summarise and reset); this file is not — it is the single source of truth the
+[`/issues` skill](../.claude/skills/issues/SKILL.md) reads back and updates.
+
+**Rule of thumb:** if it is worth remembering after this session ends, it belongs here.
+
+## How this is used
+
+- Say `/issues` in Claude Code → the skill reads this file and states the open items back,
+  grouped by priority with a one-line summary count. Nothing is mutated on a plain read.
+- `/issues add …`, `/issues done <id>`, `/issues capture`, and friends mutate the tables below.
+  The full command surface lives in the skill file.
+- Every mutation keeps this file committed so the memory survives across sessions and worktrees.
+
+## Conventions
+
+- **ID** is a monotonic `#NNN`, never reused. Allocate the next number above the current max
+  across _both_ tables (open + resolved).
+- **Pri**: `P1` (do next / blocking), `P2` (should do), `P3` (nice-to-have / revisit-when).
+- **Type**: `task` (a concrete unit of work), `rec` (a recommendation to weigh), or
+  `issue` (a defect / risk / gap).
+- **Detail / next action** is the smallest thing that would move the item forward.
+- **Source** points at where it came from: a doc, a PR (`#123`), a file:line, or `session YYYY-MM-DD`.
+- Resolving an item moves its row to **Resolved / archive** with the date and a one-line outcome —
+  rows are archived, not deleted, so the history stays auditable.
+
+<!-- issues:next-id=006 -->
+
+## Open items
+
+| ID   | Pri | Type | Summary                                          | Detail / next action                                                                                                                                                                                                                                                                                                  | Source                                                                                  | Added      |
+| ---- | --- | ---- | ------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | ---------- |
+| #001 | P2  | task | Semantic reranking still gated off               | `RAG_SEMANTIC_RERANK_ENABLED=false` from PR #901. Do not enable until the provider-backed 36/36 retrieval-quality gate **and** an ambiguity-focused canary are explicitly approved and recorded.                                                                                                                      | `docs/process-hardening.md` (Semantic reranking rollout debt); PR #901                  | 2026-07-21 |
+| #002 | P2  | task | Process-ownership fix not yet isolated on `main` | `scripts/run-eval-safe.mjs` still terminates residual repo processes broadly. Isolate the child-owned `terminateOwnedProcessTree(child.pid)` approach + the regression test proving unrelated Vitest/Playwright/Next processes survive; verify statically — do **not** start a provider-backed eval.                  | `docs/process-hardening.md` (High-priority local process ownership)                     | 2026-07-21 |
+| #003 | P2  | task | Staging tenancy release evidence outstanding     | The provider-backed A/B tenancy regression is intentionally outside local + PR gates. Run the standalone manual/nightly workflow and attach a recent green evidence artifact **before release**.                                                                                                                      | `docs/process-hardening.md`; `docs/staging-tenancy-release-evidence.md`                 | 2026-07-21 |
+| #004 | P3  | rec  | Rescope provider-gated RAG safety ideas          | If explicitly approved, rescope only the still-relevant preflight utilities/tests (answer-quality thresholds, cost-cap preflight, deep-health) against current `main`; keep live OpenAI/Supabase validation separate. Do **not** replay the stale 754-line retrieval migration/worktree wholesale.                    | `docs/process-hardening.md` (Provider-gated RAG safety ideas)                           | 2026-07-21 |
+| #005 | P3  | rec  | `finalScore` saturates at clamp ceiling          | Base + ~40 stacked boosts routinely exceed 1.0, so strong matches tie at 1.0 and order by an arbitrary `document_id` tiebreak. If ranking is ever revisited, break ties by the **pre-clamp** score rather than raising the `[0,1]` ceiling (downstream gates assume `[0,1]`). Not a defect on the current golden set. | `docs/rag-hybrid-findings-and-todo.md` P1 item 4; `src/lib/rag/clinical-search.ts:1362` | 2026-07-21 |
+
+## Resolved / archive
+
+Move resolved rows here with the resolution date and a one-line outcome. Keep them — do not delete.
+
+| ID           | Type | Summary | Outcome | Resolved |
+| ------------ | ---- | ------- | ------- | -------- |
+| _(none yet)_ |      |         |         |          |