diff --git a/.gitignore b/.gitignore
index bb92923..e3075f5 100644
--- a/.gitignore
+++ b/.gitignore
@@ -43,3 +43,4 @@ __screenshots__/
 
 # Agent runtime artifacts (regenerated per sweep, never source)
 .doc-keeper-report.md
+.opencode/cache/
diff --git a/.opencode/agent/issue-reviewer.md b/.opencode/agent/issue-reviewer.md
new file mode 100644
index 0000000..444ed0e
--- /dev/null
+++ b/.opencode/agent/issue-reviewer.md
@@ -0,0 +1,134 @@
+---
+description: Read-only issue triage agent. Loads cached GitHub issues, cross-references codebase via grep, synthesizes prioritization report. Never edits files.
+mode: subagent
+permission:
+  edit: deny
+  bash: allow
+  webfetch: allow
+---
+
+# Issue Reviewer — Issue Triage
+
+Read-only issue triage. Loads cached GitHub issues, cross-references against the codebase via grep, synthesizes a prioritization report. Does NOT edit code or files.
+
+## Inputs
+
+- `.opencode/cache/issues-review.json` — produced by `scripts/issues-review.sh`. If missing, run the script first.
+- `$ARGUMENTS` (optional) — issue number to deep-dive. Skips triage, focuses on one issue.
+
+## Process
+
+### 1. Load cache
+
+```bash
+test -f .opencode/cache/issues-review.json || bash scripts/issues-review.sh
+```
+
+If cache is stale (>1h) and issue inflow is suspected, force refresh: `bash scripts/issues-review.sh --fresh`.
+
+### 2. Per-issue grep
+
+For each open issue, extract 2–5 keywords from title + first paragraph of body. Run grep against:
+
+- `src/` (frontend)
+- `src-tauri/src/` and `src-tauri/crates/*/src/` (Rust backend)
+- `scripts/`, `.github/workflows/` (CI/infra)
+
+Use grep output to ground "Relevant Files" — never infer from project layout knowledge. If grep returns nothing relevant, say so explicitly.
+
+Skip grep for clearly non-code issues (docs, deps, CI-only).
+
+### 3. Classify each issue
+
+For each issue, determine:
+
+- **Type:** bug / feature / enhancement / question / chore
+- **Severity:**
+  - `critical` — data loss, security, build broken
+  - `high` — blocks workflow, affects many users, CI red
+  - `medium` — degrades workflow, workaround exists
+  - `low` — cosmetic, edge case
+- **App area:** editor / admin / connection / schema / ai / grid / ci-deps / docs / other
+- **Effort:**
+  - `small` — <50 LOC, single file
+  - `medium` — 1–3 files, single component/crate
+  - `large` — multi-file, design or refactor needed
+
+  Ground effort estimate in grep results (count files to touch, check crate boundaries). If ungrounded, omit rather than guess.
+
+### 4. Output format
+
+**Low volume (N ≤ 3):** skip the Summary block. Lead straight into per-issue sections.
+
+**High volume (N > 3):** include a Summary block first.
+
+```
+# Issue Triage — <YYYY-MM-DD>
+
+## Summary
+
+- N open: X bugs, Y features, Z enhancements
+- By severity: W critical, X high, Y medium, Z low
+- N issues linked to WIP PRs
+- Oldest issue: N days
+
+## Issues
+
+### #<num>: <title>
+
+- **Type / Severity / Area:** bug / high / ci-deps
+- **Age:** N days | **Labels:** bug | **Linked PRs:** #N (or "none")
+- **What it is:** <1-line summary>
+
+**Relevant files (verified via grep):**
+- `src-tauri/Cargo.toml:42` — <why>
+
+**Approach:**
+<concrete change: which file, what to add/modify. If unsure, say "needs investigation.">
+
+**Effort:** small/medium/large
+
+---
+```
+
+### 5. Recommend top 3
+
+End with prioritized recommendations. One line each:
+
+```
+## Recommended order
+
+1. **#N** — <why first>. Unblocks X.
+2. **#M** — <why second>.
+3. **#K** — <why third>.
+```
+
+If N ≤ 2, just recommend all of them.
+
+## Output style
+
+- Terse. Code paths + line numbers over prose.
+- One finding per line where possible.
+- No emojis in filenames. Severity emojis (🔴🟠🟡) OK in chat output.
+- Skip boilerplate at low N.
+- End every report with a one-line "Next action" suggestion (e.g., "Run `/review-prs` to triage open PRs.").
+
+## Memory (optional, silently skipped if unavailable)
+
+If cognee MCP is available, store a digest at the end of the report:
+
+```
+attempt: cognee_remember(data="<YYYY-MM-DD> triage: N open. Top: #X <one-liner>, #Y <one-liner>.")
+on failure: skip silently, do not abort
+```
+
+Detect cognee availability by attempting `cognee_get_client_info_json` once at start. If it errors or times out, skip all `cognee_*` calls for the session. Never block the report on memory.
+
+## Don't
+
+- Don't edit any files. Read-only agent.
+- Don't infer file paths from project layout — always grep.
+- Don't include "effort" without grounding.
+- Don't pad output for low-volume backlogs.
+- Don't recommend solutions that cross ownership lanes without flagging (frontend impl vs Rust impl vs docs).
+- Don't fetch comments for every issue — only deep-dive targets.
diff --git a/.opencode/command/reviewissues.md b/.opencode/command/reviewissues.md
new file mode 100644
index 0000000..15d5c5e
--- /dev/null
+++ b/.opencode/command/reviewissues.md
@@ -0,0 +1,74 @@
+---
+description: Triage open GitHub issues. Pulls issues + cross-references PRs, greps codebase for relevant files, prioritizes. Read-only — produces a report, not changes.
+agent: build
+---
+
+Triage open GitHub issues for SQLPilot. Pure analysis, no code changes.
+
+## What this does
+
+1. Runs `scripts/issues-review.sh` — pulls open issues + open PRs, caches to `.opencode/cache/issues-review.json`
+2. Cross-references: which issues have WIP PRs? (`Closes #N` / `Fixes #N` in PR bodies)
+3. Loads the cache, embodies `.opencode/agent/issue-reviewer.md`
+4. For each issue, greps `src/` + `src-tauri/` to ground relevant files
+5. Synthesizes per-issue: type, severity, area, approach, effort
+6. Stores digest to memory (if cognee MCP available — optional, silently skipped if not)
+7. Returns top 3 prioritized recommendations
+
+## When to invoke
+
+- Morning review: check overnight issue inflow
+- Before sprint/milestone planning
+- "What should we work on?"
+- Backlog growing and needs triage
+- After a release — verify no new critical bugs landed
+
+## Invocation
+
+```
+/reviewissues                  # full triage of all open issues
+/reviewissues 175              # deep-dive on one issue
+```
+
+Or run directly:
+
+```bash
+bash scripts/issues-review.sh             # refresh cache + print path
+bash scripts/issues-review.sh --stats     # summary stats only
+bash scripts/issues-review.sh --fresh     # force cache refresh
+```
+
+Cache is gitignored at `.opencode/cache/`. Re-runs within 1h use the cache. Override TTL with `ISSUES_REVIEW_CACHE_TTL=0` or pass `--fresh`.
+
+## Pre-flight (script enforces, aborts on fail)
+
+- `gh` CLI installed + authenticated
+- `jq` installed
+- Inside a git repo
+
+## Output
+
+Terse report. Skips Summary block at low N (≤3). Top 3 recommendations at end. No code changes.
+
+## Failure modes
+
+| Symptom                | Cause                      | Fix                                               |
+| ---------------------- | -------------------------- | ------------------------------------------------- |
+| "gh not authenticated" | gh auth expired            | `gh auth login`                                   |
+| "jq not installed"     | missing dep                | `brew install jq` / `apt install jq`              |
+| Cache missing          | first run                  | script auto-creates                               |
+| cognee MCP absent      | optional memory            | skipped silently, report still produced           |
+| Cache stale >1h        | long session, fresh issues | pass `--fresh` or set `ISSUES_REVIEW_CACHE_TTL=0` |
+
+## See also
+
+- `scripts/issues-review.sh` — data-pull script (deterministic, no LLM)
+- `.opencode/agent/issue-reviewer.md` — agent definition (embodied on invocation)
+- `/review-prs` — companion: review open PRs
+- `.opencode/AGENTS.md` — Self-Dispatch protocol
+
+## Config
+
+- `SQLPILOT_REPO_DIR` — repo location (default: pwd)
+- `ISSUES_REVIEW_CACHE_TTL` — cache TTL in sec (default: 3600)
+- Cache file: `.opencode/cache/issues-review.json` (gitignored)
diff --git a/.opencode/ops/cargo-audit-fix-plan.md b/.opencode/ops/cargo-audit-fix-plan.md
index 00ab907..b68c39e 100644
--- a/.opencode/ops/cargo-audit-fix-plan.md
+++ b/.opencode/ops/cargo-audit-fix-plan.md
@@ -30,6 +30,7 @@ Each phase lives on a sub-branch off `fix/204-cargo-audit-advisories` to isolate
 ## Test Plan
 
 After each phase:
+
 - `cargo build --workspace`
 - `cargo test --workspace`
 - `cargo clippy --workspace -- -D warnings`
diff --git a/README.md b/README.md
index 776d0f0..8386453 100644
--- a/README.md
+++ b/README.md
@@ -292,7 +292,7 @@ RUST_LOG=debug
 
 | Technology                                     | Purpose                                                |
 | ---------------------------------------------- | ------------------------------------------------------ |
-| [Vitest](https://vitest.dev/)                  | Frontend unit tests (1488 tests)                      |
+| [Vitest](https://vitest.dev/)                  | Frontend unit tests (1488 tests)                       |
 | [cargo test](https://doc.rust-lang.org/cargo/) | Rust integration tests against Docker MySQL (45 tests) |
 | [Docker](https://www.docker.com/)              | MySQL 8, MySQL 5.7, MariaDB 11 test containers         |
 
diff --git a/scripts/issues-review.sh b/scripts/issues-review.sh
new file mode 100755
index 0000000..9708af7
--- /dev/null
+++ b/scripts/issues-review.sh
@@ -0,0 +1,171 @@
+#!/usr/bin/env bash
+# scripts/issues-review.sh
+#
+# Pull all open GitHub issues + cross-referenced PRs into a single cache file.
+# Feeds the issue-reviewer agent. Pure bash + gh + jq. No LLM calls.
+#
+# Cache schema (.opencode/cache/issues-review.json):
+# {
+#   "repo": "owner/name",
+#   "generated_at": "2026-07-19T12:34:56Z",
+#   "open_issues": [{number, title, body, labels, author, createdAt, updatedAt, comment_count}, ...],
+#   "open_prs": [{number, title, body, headRefName, author, createdAt}, ...],
+#   "linked_prs": [{pr, title, refs: [issue_numbers]}, ...],
+#   "summary": {open_count, by_label, oldest_age_days, linked_to_open_pr}
+# }
+#
+# Usage:
+#   scripts/issues-review.sh           # use cache if <TTL, else refresh
+#   scripts/issues-review.sh --fresh   # force refresh
+#   scripts/issues-review.sh --stats   # print summary only, don't refresh
+#
+# Exit codes:
+#   0  - success (cache fresh or refreshed)
+#   1  - pre-flight failure (gh/jq missing, not authed)
+#   2  - fetch failed
+#   3  - cache write failed
+
+set -euo pipefail
+
+REPO_DIR="${SQLPILOT_REPO_DIR:-$(pwd)}"
+CACHE_DIR="${REPO_DIR}/.opencode/cache"
+CACHE_FILE="${CACHE_DIR}/issues-review.json"
+CACHE_TTL_SEC="${ISSUES_REVIEW_CACHE_TTL:-3600}"
+LOG_PREFIX="[issues-review]"
+LIMIT=500
+
+log()  { printf '%s %s %s\n' "$LOG_PREFIX" "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$*"; }
+err()  { printf '%s ERROR: %s\n' "$LOG_PREFIX" "$*" >&2; }
+die()  { err "$*"; exit "${2:-1}"; }
+
+usage() {
+  sed -n '2,20p' "$0"
+  exit 0
+}
+
+# --- Parse args ---
+FRESH=false
+STATS_ONLY=false
+for arg in "$@"; do
+  case "$arg" in
+    --fresh) FRESH=true ;;
+    --stats) STATS_ONLY=true ;;
+    --help|-h) usage ;;
+    -*) die "Unknown flag: $arg" 1 ;;
+    *) die "Unexpected arg: $arg" 1 ;;
+  esac
+done
+
+# --- Pre-flight ---
+[ -d "$REPO_DIR/.git" ] || die "Not a git repo: $REPO_DIR" 1
+command -v gh >/dev/null || die "gh CLI not installed" 1
+command -v jq >/dev/null || die "jq not installed (brew install jq / apt install jq)" 1
+gh auth status >/dev/null 2>&1 || die "gh not authenticated (run: gh auth login)" 1
+
+mkdir -p "$CACHE_DIR"
+
+# --- Cache check (unless --fresh) ---
+if [ "$FRESH" = false ] && [ -f "$CACHE_FILE" ]; then
+  MTIME=$(stat -c %Y "$CACHE_FILE" 2>/dev/null || stat -f %m "$CACHE_FILE")
+  AGE=$(($(date +%s) - MTIME))
+  if [ "$AGE" -lt "$CACHE_TTL_SEC" ]; then
+    log "Cache fresh (${AGE}s old, TTL ${CACHE_TTL_SEC}s). Use --fresh to force refresh."
+    if [ "$STATS_ONLY" = true ]; then
+      jq -r '
+        "Open issues: \(.summary.open_count)",
+        "By label: \(.summary.by_label | tojson)",
+        "Linked to open PR: \(.summary.linked_to_open_pr)",
+        "Oldest issue age: \(.summary.oldest_age_days // "n/a") days"
+      ' "$CACHE_FILE"
+    else
+      printf '%s\n' "$CACHE_FILE"
+    fi
+    exit 0
+  fi
+  log "Cache stale (${AGE}s > TTL ${CACHE_TTL_SEC}s). Refreshing."
+fi
+
+# --- Fetch ---
+log "Fetching open issues + PRs (limit ${LIMIT})..."
+REPO="$(gh repo view --json nameWithOwner -q .nameWithOwner)" || die "gh repo view failed" 2
+
+OPEN_ISSUES_JSON="$(gh issue list --state open --limit "$LIMIT" \
+  --json number,title,body,labels,author,createdAt,updatedAt,comments 2>&1)" \
+  || die "gh issue list failed" 2
+
+OPEN_PRS_JSON="$(gh pr list --state open --limit "$LIMIT" \
+  --json number,title,body,headRefName,author,createdAt 2>&1)" \
+  || die "gh pr list failed" 2
+
+log "Fetched $(echo "$OPEN_ISSUES_JSON" | jq 'length') issues + $(echo "$OPEN_PRS_JSON" | jq 'length') PRs."
+
+# --- Build linked_prs: for each open PR, extract #N refs from body, filter to open issues ---
+LINKED_PRS_JSON="$(echo "$OPEN_PRS_JSON" | jq --argjson issues "$OPEN_ISSUES_JSON" '
+  ($issues | map(.number)) as $nums |
+  map(
+    . as $pr |
+    ([((.body // "") | scan("#(\\d+)"; "g")) | .[] | .[0] | tonumber] | unique) as $refs |
+    ($refs | map(select(. as $n | $nums | index($n))) ) as $matches |
+    select($matches | length > 0) |
+    {pr: .number, title: .title, refs: $matches}
+  )
+')"
+
+# --- Assemble final cache ---
+FINAL_JSON="$(jq -n \
+  --arg repo "$REPO" \
+  --arg generated_at "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
+  --argjson issues "$OPEN_ISSUES_JSON" \
+  --argjson prs "$OPEN_PRS_JSON" \
+  --argjson linked "$LINKED_PRS_JSON" \
+  '
+  {
+    repo: $repo,
+    generated_at: $generated_at,
+    open_issues: $issues,
+    open_prs: $prs,
+    linked_prs: $linked,
+    open_issues: ($issues | map(. + {comment_count: (.comments | length)})),
+    summary: {
+      open_count: ($issues | length),
+      by_label: (
+        $issues
+        | map(.labels[].name)
+        | group_by(.)
+        | map({name: .[0], value: length})
+        | from_entries
+      ),
+      oldest_age_days: (
+        if ($issues | length) > 0
+        then ((now - ($issues | min_by(.createdAt).createdAt | fromdateiso8601)) / 86400 | floor)
+        else null
+        end
+      ),
+      linked_to_open_pr: (
+        $linked | map(.refs[]) | unique | length
+      )
+    }
+  }
+')"
+
+# --- Write cache atomically ---
+TMP="$(mktemp "${CACHE_DIR}/issues-review.XXXXXX.json")"
+trap 'rm -f "$TMP"' EXIT
+echo "$FINAL_JSON" > "$TMP" || die "Cache write failed: $TMP" 3
+mv "$TMP" "$CACHE_FILE" || die "Cache mv failed: $CACHE_FILE" 3
+trap - EXIT
+
+log "Cached → $CACHE_FILE"
+
+# --- Output ---
+if [ "$STATS_ONLY" = true ]; then
+  jq -r '
+    "Open issues: \(.summary.open_count)",
+    "By label: \(.summary.by_label | tojson)",
+    "Linked to open PR: \(.summary.linked_to_open_pr)",
+    "Oldest issue age: \(.summary.oldest_age_days // "n/a") days"
+  ' "$CACHE_FILE"
+else
+  printf '%s\n' "$CACHE_FILE"
+fi
+exit 0