A sidebar on Ctrl+B: drill's first plugin, clickable to the last gesture

[jainal09]

Stacked on #17 (feat/macos-cmd) — merge that first; this PR retargets main cleanly after.

What this is

An optional file-explorer sidebar on Ctrl+B / Cmd+B — drill's first and only plugin-powered feature, built to stay an exception:

• nvim-tree draws the tree, nvzone/volt + nvzone/menu draw the right-click menu — three checkouts pinned to exact SHAs by the new vendor.sh (a pin table and a loop, not a plugin manager).
• Nothing loads at startup. The checkouts aren't on the runtime path until the first toggle; never press Ctrl+B and you run the config drill always had.
• The same branch seals the packpath leak: user plugins in ~/.local/share/nvim/site/pack/*/start loaded inside drill despite the "no plugins" promise (measured). Both site dirs now come off runtimepath and packpath, with a suite assert.

The gestures (all mouse, all pty-tested)

• One click opens a file (never into the REPL — window_picker excludes terminals) or folds a folder.
• [+ File] / [+ Folder] — real clickable winbar buttons (%@handler@ regions). The folder button does mkdir -p, not the trailing-slash typing convention.
• Right-click menu — plain-ASCII items (New file/folder, Open, Open in split, Rename, Cut, Copy, Paste, Delete), global-but-late mapping (mouse events route to the focused buffer's maps, so tree-local could never fire); outside the tree the stock right-click is re-fed untouched.
• Ctrl+click multi-select (marks, rendered *), and drag-and-drop: drop a file on a folder to move it in; drag a marked row and the whole selection moves. A one-cell jitter is a click, not a drag — same trackpad rule as the rest of drill.

Bugs found on the way

• A mouse press from insert focuses the tree while staying in insert (measured: mode "i", filetype NvimTree), landing releases in a mode no tree mapping owns — every click silently ate itself. The tree now forces Normal on entry.
• <D-b> stops being the docs' example of a floored chord; the floor test re-spells with <D-g>.

Tests & plumbing

• New tests/suite_tree.sh + tests/tree_drive.py: 45 cases on a real pty — both toggle chords (Cmd+B as raw CSI-u bytes), focus that never leaves your code, toolbar clicks all the way to files on disk, a menu item clicked by its float coordinates, marked-pair drag moves, and the 3-window REPL layout.
• tests/run.sh gains a vendor.sh preflight (CI needs zero workflow changes); gate grows 558 → 719 cases, all green locally (2 full runs).
• install.sh: manifest + vendor fetch (FEATURE-level; offline installs lose only the sidebar), nvim gate 0.9 → 0.10 (what the pinned nvim-tree requires). Verified in a sandboxed HOME.
• Docs: README section, KEYS.md gestures, new docs/explorer.md, honesty pass on every "no plugins" claim.

🤖 Generated with Claude Code

---

Summary by cubic

Adds an optional, vendored file-explorer sidebar on Ctrl+B/Cmd+B that loads only on first toggle. Old behavior: no plugins and the sidebar (when present) rooted at the shell’s working directory with possible user plugin leakage. New behavior: one opt‑in plugin (nvim-tree.lua + nvzone/volt + nvzone/menu) loads lazily, the tree roots at the current file’s folder on open, it keeps the existing root when toggled from the REPL or a scratch buffer, filesystem watchers stay disabled to avoid EMFILE, and both runtimepath and packpath seals block third‑party loads. Hiding the REPL now returns focus to a real typing buffer instead of the tree.

• Focus stays in the file window; single‑click opens/folds; real winbar “[+ File]” / “[+ Folder]”; right‑click menu only in the tree; Ctrl+click marks; drag‑and‑drop moves (marked sets move together); jitter is a click; entering the tree forces Normal; window picker excludes terminals. Netrw roots the tree at the directory it lists. Docs call out iTerm2’s right‑click Pointer binding that must be removed.
• Code review: explorer.lua lazy setup, drag/drop, menu, toolbar, rooting and watcher settings; nvimrc.lua path sealing, Ctrl+B/Cmd+B toggles, and repl_hide focus fix; vendor.sh pinned checkouts; install.sh runs vendor.sh, adds a git dep, and enforces Neovim 0.10+; tests/run.sh preflights vendor.sh; gate is 729 cases with 55 real‑PTY sidebar tests; docs updated.

Rollout

• Required: run ./vendor.sh after pulling; ensure Neovim 0.10+ and git are installed.
• macOS: enable Cmd forwarding; in iTerm2 remove the default right‑click Pointer binding so the tree menu works.
• No change if the sidebar is never toggled; netrw remains for directory listing.

^{Written for commit 9a1a3c2. Summary will update on new commits.}

Greptile Summary

The PR adds an optional, lazily loaded file-explorer sidebar and vendors its pinned plugin dependencies.

• Adds mouse-driven tree navigation, creation, context-menu, selection, and drag-and-drop behavior.
• Seals user site directories out of Neovim runtime and package paths.
• Extends installation, documentation, and PTY-based test coverage for the sidebar.

Confidence Score: 3/5

The PR is not yet safe to merge because vendor refresh failures can remove working sidebar dependencies and the installer rejects a still-documented supported Neovim version.

The existing vendor replacement path deletes each nonmatching checkout before fetching its replacement, and the installation contract still disagrees between the hard Neovim 0.10 gate and README’s 0.9+ prerequisite.

Files Needing Attention: vendor.sh, install.sh, README.md

Important Files Changed

Filename	Overview
explorer.lua	Implements lazy sidebar setup, tree rooting, mouse gestures, toolbar actions, context menus, and drag-and-drop.
nvimrc.lua	Seals user plugin paths and adds lazy sidebar toggle mappings while preserving the isolated editor setup.
vendor.sh	Fetches three exact plugin revisions, but the previously reported destructive replacement sequence remains.
install.sh	Installs sidebar assets and dependencies, but its Neovim hard gate remains inconsistent with the documented prerequisite.
tests/suite_tree.sh	Adds real-PTY coverage for sidebar toggles, focus, mouse interactions, filesystem operations, and REPL layouts.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  K[Ctrl+B or Cmd+B] --> L[Load explorer.lua]
  L --> V[Add pinned vendor checkouts to runtimepath]
  V --> T[Initialize nvim-tree and menu]
  T --> U[Mouse-driven sidebar]
  I[install.sh or tests/run.sh] --> F[vendor.sh]
  F --> V

Loading

_{Reviews (3): Last reviewed commit: "Ctrl+E out of the REPL means the code, n..." | Re-trigger Greptile}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 64be8e83-84b6-4f38-b598-b797897b2618

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[greptile-apps]

Failed fetch destroys working checkout

When an existing checkout differs from the configured pin and the replacement fetch fails, vendor.sh deletes the working checkout before downloading its replacement, leaving an incomplete vendor directory and disabling the entire sidebar.

Knowledge Base Used: Installation and launch flow

[cubic-dev-ai]

11 issues found across 17 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="README.md">

<violation number="1" location="README.md:116">
P3: The README still tells users 'Requires Neovim 0.9+', but this PR bumps install.sh's Neovim gate to 0.10+ (the sidebar's pinned plugins need 0.10). Update the Install details line to match so the documented minimum matches the enforced minimum.</violation>
</file>

<file name="tests/suite_tree.sh">

<violation number="1" location="tests/suite_tree.sh:45">
P3: Running with a filter that matches nothing (e.g. `./suite_tree.sh clik` with a typo) prints `PASS=0 FAIL=0` and exits 0, so the run is reported as green even though no case executed. When FILTER is non-empty, guard the success path on at least one matched case (or print the filter in the summary) so a mistyped filter cannot silently pass.</violation>
</file>

<file name="docs/explorer.md">

<violation number="1" location="docs/explorer.md:33">
P3: The doc promises that `Ctrl+E` and `Ctrl+R` "do nothing" inside the tree, but the global mappings still act there: `repl_show` and `run_once` only gate on `source_state()`, which, because the NvimTree buffer fails `typing_buffer()` (it is not modifiable and is not the netrw listing), simply returns the stale non-nil `last_file`. So once any file has been opened, pressing `Ctrl+E` in the tree focuses/opens the interpreter and `Ctrl+R` runs the file, neither of which is "nothing" — the same applies in the netrw listing the doc cites as the comparison. The suite only asserts "no wedge / no extra window" (tests/tree_drive.py), not a true no-op, so the tests don't back the doc claim. Either narrow the sentence to what actually holds (only `Ctrl+S`, which `save()` guards on `typing_buffer()`, is a no-op) or add a buffer-type guard to `repl_show`/`run_once` if the intent is genuinely inert keys in the tree.</violation>

<violation number="2" location="docs/explorer.md:57">
P3: install.sh now hard-requires Neovim 0.10+ (HARD abort), but README.md still states "Requires Neovim 0.9+" and this new line implies the rest of drill runs fine on 0.9. Align the minimum version in README.md (and soften this bullet) so users don't rely on 0.9 support that the installer refuses.</violation>
</file>

<file name="install.sh">

<violation number="1" location="install.sh:187">
P2: When apt's candidate is Neovim 0.9.x, this new gate rejects it, but `nvim_pkg` still offers `neovim` because its filter stops at 0.8. The installer therefore suggests an apt command that installs the same rejected version, then aborts; extend the candidate filter through 0.9.</violation>
</file>

<file name="vendor.sh">

<violation number="1" location="vendor.sh:35">
P2: When a pinned checkout is dirty, this fast path skips cleanup, so modified or extra plugin files remain loaded by the sidebar. Check the worktree is clean before declaring the pin ready, or recreate it.</violation>

<violation number="2" location="vendor.sh:40">
P2: When a pin update cannot fetch, this deletes the existing checkout first and leaves the sidebar unavailable. Fetch into a temporary checkout and replace the old directory only after checkout and verification succeed.</violation>
</file>

<file name="explorer.lua">

<violation number="1" location="explorer.lua:204">
P2: The stale `<LeftMouse>` auto-close map left by nvzone/menu is only deleted at the start of the *next* `tree_menu()` call. When a menu closes by item click (the common path), the plugin's auto_close map never deletes itself, so a leftover global `<LeftMouse>` mapping stays installed in the file buffer until the user right-clicks again. It re-feeds the built-in press so it usually doesn't visibly break anything, but the cleanup is in the wrong place. Clear it in the menu's after_close path (or have the store delete it as soon as the menu closes) instead of relying on the next open.</violation>

<violation number="2" location="explorer.lua:261">
P1: When a drag leaves the tree and releases over a file window, the tree’s buffer-local `<LeftRelease>` mapping does not clear `drag_src`. The next ordinary tree click can therefore call `drop` and move the stale file; clear drag state on release or window leave regardless of the receiving window.</violation>
</file>

<file name="nvimrc.lua">

<violation number="1" location="nvimrc.lua:11">
P1: Users with `~/.config/nvim/plugin` or `after/plugin` files can still execute those plugins in drill, despite the new no-plugin seal. Remove `stdpath("config")` and its `/after` entry from both paths, along with any other non-bundled site entries, or disable startup plugin loading.</violation>

<violation number="2" location="nvimrc.lua:1341">
P2: When the sidebar is unavailable, pressing Ctrl+B from insert mode leaves the user in Normal mode because `stopinsert` runs before the error return. Check the module error before stopping insert and call `type_here()` on this failure path.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P1: When a drag leaves the tree and releases over a file window, the tree’s buffer-local <LeftRelease> mapping does not clear drag_src. The next ordinary tree click can therefore call drop and move the stale file; clear drag state on release or window leave regardless of the receiving window.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At explorer.lua, line 261:

<comment>When a drag leaves the tree and releases over a file window, the tree’s buffer-local `<LeftRelease>` mapping does not clear `drag_src`. The next ordinary tree click can therefore call `drop` and move the stale file; clear drag state on release or window leave regardless of the receiving window.</comment>

<file context>
@@ -0,0 +1,313 @@
+
+local function tree_click()
+  local mp = vim.fn.getmousepos()
+  local src = drag_src
+  drag_src = nil
+  if mp.line == 0 then return end
</file context>

[cubic-dev-ai]

P1: Users with ~/.config/nvim/plugin or after/plugin files can still execute those plugins in drill, despite the new no-plugin seal. Remove stdpath("config") and its /after entry from both paths, along with any other non-bundled site entries, or disable startup plugin loading.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At nvimrc.lua, line 11:

<comment>Users with `~/.config/nvim/plugin` or `after/plugin` files can still execute those plugins in drill, despite the new no-plugin seal. Remove `stdpath("config")` and its `/after` entry from both paths, along with any other non-bundled site entries, or disable startup plugin loading.</comment>

<file context>
@@ -1,6 +1,18 @@
+-- packages load from 'packpath', which still held it -- measured, a plugin in
+-- ~/.local/share/nvim/site/pack/x/start/ loaded inside drill. All three
+-- doors close here, so "no plugins" means the machine's plugins too.
+for _, p in ipairs({ vim.fn.stdpath("data") .. "/site",
+                     vim.fn.stdpath("data") .. "/site/after" }) do
+  vim.opt.runtimepath:remove(p)
</file context>

[cubic-dev-ai]

P2: When apt's candidate is Neovim 0.9.x, this new gate rejects it, but nvim_pkg still offers neovim because its filter stops at 0.8. The installer therefore suggests an apt command that installs the same rejected version, then aborts; extend the candidate filter through 0.9.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At install.sh, line 187:

<comment>When apt's candidate is Neovim 0.9.x, this new gate rejects it, but `nvim_pkg` still offers `neovim` because its filter stops at 0.8. The installer therefore suggests an apt command that installs the same rejected version, then aborts; extend the candidate filter through 0.9.</comment>

<file context>
@@ -184,10 +184,10 @@ check_deps() {
     maj="$(printf '%s' "${cand%%.*}" | tr -cd '0-9')"
     rest="${cand#*.}"; min="$(printf '%s' "${rest%%.*}" | tr -cd '0-9')"
-    if [ "${maj:-0}" -eq 0 ] && [ "${min:-0}" -lt 9 ]; then
+    if [ "${maj:-0}" -eq 0 ] && [ "${min:-0}" -lt 10 ]; then
       NOTES="$NOTES
-    - neovim $cand is too old, need 0.9+ (the config is Lua and uses nvim_win_hide).
</file context>

[cubic-dev-ai]

P2: When a pin update cannot fetch, this deletes the existing checkout first and leaves the sidebar unavailable. Fetch into a temporary checkout and replace the old directory only after checkout and verification succeed.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At vendor.sh, line 40:

<comment>When a pin update cannot fetch, this deletes the existing checkout first and leaves the sidebar unavailable. Fetch into a temporary checkout and replace the old directory only after checkout and verification succeed.</comment>

<file context>
@@ -0,0 +1,49 @@
+    continue
+  fi
+  echo "vendor.sh: fetching $name @ ${sha:0:12}"
+  rm -rf "$dest"
+  mkdir -p "$dest"
+  git -C "$dest" init -q
</file context>

[cubic-dev-ai]

P2: When a pinned checkout is dirty, this fast path skips cleanup, so modified or extra plugin files remain loaded by the sidebar. Check the worktree is clean before declaring the pin ready, or recreate it.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At vendor.sh, line 35:

<comment>When a pinned checkout is dirty, this fast path skips cleanup, so modified or extra plugin files remain loaded by the sidebar. Check the worktree is clean before declaring the pin ready, or recreate it.</comment>

<file context>
@@ -0,0 +1,49 @@
+  [ -z "$name" ] && continue
+  dest="$VENDOR/$name"
+  have="$(git -C "$dest" rev-parse HEAD 2>/dev/null || true)"
+  if [ "$have" = "$sha" ]; then
+    echo "vendor.sh: $name already at ${sha:0:12}"
+    continue
</file context>

[cubic-dev-ai]

P2: The stale <LeftMouse> auto-close map left by nvzone/menu is only deleted at the start of the next tree_menu() call. When a menu closes by item click (the common path), the plugin's auto_close map never deletes itself, so a leftover global <LeftMouse> mapping stays installed in the file buffer until the user right-clicks again. It re-feeds the built-in press so it usually doesn't visibly break anything, but the cleanup is in the wrong place. Clear it in the menu's after_close path (or have the store delete it as soon as the menu closes) instead of relying on the next open.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At explorer.lua, line 204:

<comment>The stale `<LeftMouse>` auto-close map left by nvzone/menu is only deleted at the start of the *next* `tree_menu()` call. When a menu closes by item click (the common path), the plugin's auto_close map never deletes itself, so a leftover global `<LeftMouse>` mapping stays installed in the file buffer until the user right-clicks again. It re-feeds the built-in press so it usually doesn't visibly break anything, but the cleanup is in the wrong place. Clear it in the menu's after_close path (or have the store delete it as soon as the menu closes) instead of relying on the next open.</comment>

<file context>
@@ -0,0 +1,313 @@
+  -- menu leaves its auto-close <LeftMouse> map behind when a menu is closed
+  -- by an item click rather than an outside click; clear the stale one so
+  -- it cannot stack
+  pcall(vim.keymap.del, "n", "<LeftMouse>")
+  require("menu").open(menu_items(), { mouse = true, border = true })
+end
</file context>

[cubic-dev-ai]

P3: The README still tells users 'Requires Neovim 0.9+', but this PR bumps install.sh's Neovim gate to 0.10+ (the sidebar's pinned plugins need 0.10). Update the Install details line to match so the documented minimum matches the enforced minimum.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At README.md, line 116:

<comment>The README still tells users 'Requires Neovim 0.9+', but this PR bumps install.sh's Neovim gate to 0.10+ (the sidebar's pinned plugins need 0.10). Update the Install details line to match so the documented minimum matches the enforced minimum.</comment>

<file context>
@@ -109,6 +113,22 @@ t -k      # stop
 
 Non-blocking, with a sound and a desktop notification at zero.
 
+## `Ctrl+B` — a file sidebar, when you want one
+
+The one optional extra: press `Ctrl+B` (or `Cmd+B` on a Mac that forwards it,
</file context>

[cubic-dev-ai]

P3: Running with a filter that matches nothing (e.g. ./suite_tree.sh clik with a typo) prints PASS=0 FAIL=0 and exits 0, so the run is reported as green even though no case executed. When FILTER is non-empty, guard the success path on at least one matched case (or print the filter in the summary) so a mistyped filter cannot silently pass.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At tests/suite_tree.sh, line 45:

<comment>Running with a filter that matches nothing (e.g. `./suite_tree.sh clik` with a typo) prints `PASS=0 FAIL=0` and exits 0, so the run is reported as green even though no case executed. When FILTER is non-empty, guard the success path on at least one matched case (or print the filter in the summary) so a mistyped filter cannot silently pass.</comment>

<file context>
@@ -0,0 +1,59 @@
+
+PASS=0; FAIL=0
+declare -a BAD=()
+while IFS=$'\t' read -r verdict name got; do
+  case "$verdict" in PASS|FAIL) ;; *) continue ;; esac
+  if [ -n "$FILTER" ] && [[ "$name" != *"$FILTER"* ]]; then continue; fi
</file context>

[cubic-dev-ai]

P3: The doc promises that Ctrl+E and Ctrl+R "do nothing" inside the tree, but the global mappings still act there: repl_show and run_once only gate on source_state(), which, because the NvimTree buffer fails typing_buffer() (it is not modifiable and is not the netrw listing), simply returns the stale non-nil last_file. So once any file has been opened, pressing Ctrl+E in the tree focuses/opens the interpreter and Ctrl+R runs the file, neither of which is "nothing" — the same applies in the netrw listing the doc cites as the comparison. The suite only asserts "no wedge / no extra window" (tests/tree_drive.py), not a true no-op, so the tests don't back the doc claim. Either narrow the sentence to what actually holds (only Ctrl+S, which save() guards on typing_buffer(), is a no-op) or add a buffer-type guard to repl_show/run_once if the intent is genuinely inert keys in the tree.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At docs/explorer.md, line 33:

<comment>The doc promises that `Ctrl+E` and `Ctrl+R` "do nothing" inside the tree, but the global mappings still act there: `repl_show` and `run_once` only gate on `source_state()`, which, because the NvimTree buffer fails `typing_buffer()` (it is not modifiable and is not the netrw listing), simply returns the stale non-nil `last_file`. So once any file has been opened, pressing `Ctrl+E` in the tree focuses/opens the interpreter and `Ctrl+R` runs the file, neither of which is "nothing" — the same applies in the netrw listing the doc cites as the comparison. The suite only asserts "no wedge / no extra window" (tests/tree_drive.py), not a true no-op, so the tests don't back the doc claim. Either narrow the sentence to what actually holds (only `Ctrl+S`, which `save()` guards on `typing_buffer()`, is a no-op) or add a buffer-type guard to `repl_show`/`run_once` if the intent is genuinely inert keys in the tree.</comment>

<file context>
@@ -0,0 +1,60 @@
+handling lives by, so nothing ever moves by accident.
+
+Inside the tree, drill's file keys say no the same way they do in the
+directory listing: `Ctrl+S`, `Ctrl+E`, `Ctrl+R` do nothing there.
+
+## What it costs when you don't use it
</file context>

[cubic-dev-ai]

P3: install.sh now hard-requires Neovim 0.10+ (HARD abort), but README.md still states "Requires Neovim 0.9+" and this new line implies the rest of drill runs fine on 0.9. Align the minimum version in README.md (and soften this bullet) so users don't rely on 0.9 support that the installer refuses.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At docs/explorer.md, line 57:

<comment>install.sh now hard-requires Neovim 0.10+ (HARD abort), but README.md still states "Requires Neovim 0.9+" and this new line implies the rest of drill runs fine on 0.9. Align the minimum version in README.md (and soften this bullet) so users don't rely on 0.9 support that the installer refuses.</comment>

<file context>
@@ -0,0 +1,60 @@
+- *"sidebar plugins are not fetched"* — run `./vendor.sh` in your drill
+  directory (`~/drill` unless you moved it). `install.sh` does this for you;
+  a `git pull` upgrade does not.
+- *"the sidebar needs nvim 0.10+"* — everything else in drill still works;
+  the pinned plugins are what need 0.10.
+- `Cmd+B` does nothing — your terminal is not forwarding Cmd. One setting:
</file context>

[codescene-access]

Gates Failed
New code is healthy (1 new file with code health below 10.00)
Enforce critical code health rules (1 file with Bumpy Road Ahead)
Enforce advisory code health rules (1 file with Complex Method)

Our agent can fix these. Install it.

Gates Passed
1 Quality Gates Passed

Reason for failure

New code is healthy	Violations	Code Health Impact
tree_drive.py	2 rules	8.49	Suppress

Enforce critical code health rules	Violations	Code Health Impact
tree_drive.py	1 critical rule	8.49	Suppress

Enforce advisory code health rules	Violations	Code Health Impact
tree_drive.py	1 advisory rule	8.49	Suppress

See analysis details in CodeScene

Quality Gate Profile: Clean Code Collective
Install CodeScene MCP: safeguard and uplift AI-generated code. Catch issues early with our IDE extension and CLI tool.