@satanwagen/reviewkit 0.1.4 → 0.1.5

package/CHANGELOG.md

@@ -1,5 +1,132 @@
 # Changelog
 
+## 0.1.5 — 2026-09-02
+
+Fixes the "status won't change" bug for good.
+
+- **Root cause:** the browser session is per origin (all pages), rooms are per
+  page, and on every connect the client uploaded its whole session into the
+  current room. Each item therefore existed in several rooms; a status change
+  updated one copy and the others stayed stale, so the web and the terminal
+  showed a mix.
+- **Client:** on connect only items of the current page are re-uploaded.
+- **Server (deployed):** items live in the room of their own `route`: an add
+  for another page is redirected there (an already-known id is confirmed,
+  never duplicated), and an update / status / delete for an id the current
+  room does not hold is applied in the sibling room that does — the sender
+  still gets the echo. On load, items stored in the wrong room are moved to
+  their own room ("rehomed"), so old duplicates disappear by themselves.
+- **CLI (0.1.12):** shows each item once even if a stale copy still exists,
+  preferring the copy in its own page's room.
+
+## 0.1.12 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- Each item is listed once even when an old duplicate lingers in another
+  room (see 0.1.5 of the main package); mutations target the right room.
+- First `e` (edit prompt) without `$EDITOR` offers to install **micro**, a
+  terminal editor with VS Code keys (Ctrl+S save, Ctrl+Q quit, Ctrl+Z undo,
+  mouse): `y` installs (Homebrew on macOS, otherwise getmic.ro into
+  `~/.local/bin`) and uses it from then on, `n` sticks with nano. The choice
+  is remembered; `EDITOR=…` in the shell always wins. micro is preferred
+  automatically when already installed.
+
+## 0.1.11 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- Prompt editing (`e`) uses `$VISUAL` / `$EDITOR`, else **nano** when
+  installed, else vi; the terminal shows how to save and leave before the
+  editor opens (nano: Ctrl+O, Enter, Ctrl+X · vi: Esc `:wq`). GUI editors
+  (code, cursor, subl, zed) get `--wait` so the run waits for you.
+
+## 0.1.10 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- **Follow-ups.** `r` on a finished run opens a reply line; what you type goes
+  to the same agent session (`--resume <session id>`) as a new run marked
+  `#4↩2`, with the original items named, so "also make the button darker"
+  lands in the right context. Works for Claude Code; Cursor gets the same
+  flag best-effort.
+- **Prompt review.** Before anything is sent you see the exact prompt: `⏎`
+  sends, `n` adds a one-line note, `e` opens it in `$EDITOR` for real edits,
+  `←` goes back to the model.
+- **Run view.** Agent runs are cards (colour bar, `WORK` / `DONE` / `FAIL`
+  badge, agent, model, items, elapsed, last step). `Tab` focuses the cards,
+  `⏎` opens a run: its items with live statuses, model, turns, cost, session,
+  the full scrollable log and a follow-up line at the bottom (`⏎` sends into
+  the same session, `y` copies the log, `K` stop, `X` close).
+- **Inline only.** The full-screen interactive agent mode is gone; agents
+  always work inside the view.
+
+## 0.1.9 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- **Parallel agents.** Up to 10 runs at once, each on its own items (an item
+  already being worked on is skipped). One line per run in the pane with a
+  `WORK` / `DONE` / `FAIL` badge, agent, model, item count, elapsed time and
+  the latest step; `Tab` focuses a run, `K` stops it, `L` shows its log, `X`
+  closes a finished one. Finished runs show `✓ n/m applied · turns · $`.
+- **Leftovers closed.** When a run ends OK, items the agent forgot to mark are
+  set to `applied` on its behalf (logged); the agent prompt now requires a
+  status call per item.
+- **Fix:** status changes from the terminal silently did nothing for items
+  whose route has a trailing slash (`/about/` vs room `/about`): the change
+  went to the wrong room. Routes are compared normalised now, so it reaches
+  the room and every browser.
+
+## 0.1.8 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- **Cursor support.** The agent picker offers every agent CLI found on PATH:
+  Claude Code (`claude`) and Cursor (`cursor-agent`, print mode with
+  `--force`, MCP from `~/.cursor/mcp.json` which `reviewkit setup` writes).
+  Two-step picker: agent → model, last choice remembered per agent.
+- **Bulk hand-over.** `space` selects items (◆), `a` hands the selection (or
+  all open items when nothing is selected) to the agent; `1`–`4` change the
+  status of every selected item at once.
+- **Agent finish = status change.** Items the agent worked on but did not
+  mark itself become `accepted` when the run ends OK.
+- **Filters that exist.** `f` opens a filter panel (environment, status, type,
+  priority, page — arrows to move and change), `/` is a live text search
+  (id, text, author, note, selector); active filters are chips in the header,
+  `F` clears. `s` cycles status, `o` cycles environment.
+- **No accidental quits.** `q` asks for a second `q`; Ctrl+C quits at once.
+  `esc` clears the selection, then the filters.
+- **Redrawn table and footer.** Type glyphs (✎ ▣ ❝ ✕ ⇄ ◐), status dots, column
+  separators, key chips in the footer, a shorter header (name · environments
+  · who is online · counts). Long descriptive hints are gone.
+- **No flicker.** Frames are painted line by line with erase-to-end-of-line
+  and redraws are coalesced (max one per 16 ms) instead of clearing the
+  screen for every update.
+
+## 0.1.7 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- **Agent works inside the live view.** `a` (all open items) or Enter → "Apply
+  with claude here" opens a picker — model per task: Fable 5.1, Opus, Sonnet,
+  Haiku or the Claude Code default (last pick preselected) — then runs
+  Claude Code in print mode with the reviewkit MCP attached and streams its
+  steps into an agent pane under the list (`▸ Edit page.tsx`, `it-3f9a →
+  applied`, cost and turns at the end). Items being worked on spin; statuses
+  flip live for reviewers too. `K` stops, `L` shows the full log, `X` closes
+  the pane, `A` / menu `C` still open full-screen interactive Claude.
+- Detail strip under the list: selector, tag, text snippet and the request of
+  the selected item.
+- Edits are auto-accepted (`--permission-mode acceptEdits`), reviewkit MCP
+  tools pre-allowed; `REVIEWKIT_AGENT_PERMISSIONS=bypassPermissions` for
+  trusted repos, `REVIEWKIT_AGENT` to swap the binary.
+
+## 0.1.6 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- **Environments.** One repo usually runs on several origins (localhost, a
+  dev domain, production); a project now carries a list of them. Pairing from
+  another environment with the same token **adds** its origin instead of
+  replacing the file; `reviewkit env` lists them, `env add <url>` / `env rm`
+  edit them. The live view shows all environments at once (a `where` column
+  with the host) and `o` switches between them; `list`/`watch` take
+  `--env <origin>`; MCP `reviewkit_list` accepts `origin` and tags items with
+  theirs. Items keep non-enumerable origin/route tags so exports stay clean.
+
+## 0.1.5 — 2026-09-02 (`@satanwagen/reviewkit-cli` only)
+
+- CLI: a `.reviewkit/emblema-sync.json` or `project.json` still pointing at
+  the retired `agropolio.fucking.style` host resolves to the default public
+  server instead of a dead URL. The main package stays at 0.1.4.
+
 ## 0.1.4 — 2026-09-02
 
 ### Added

package/README.md

@@ -312,7 +312,9 @@ reviewkit          # first run pairs the repo (one click in the review panel), t
 
 **Pairing is one click, no copy-paste.** Bare `reviewkit` in an unpaired repo runs `setup` first; it detects the site's `token` (and a custom `syncToken` / `syncUrl`, if any) in the repo, opens a loopback port and waits. In the browser open the review panel → gear → **Connect terminal / agent…** → **Connect**: the panel finds the waiting terminal and pushes the connection into `<repo>/.reviewkit/project.json` (gitignored). `setup` also registers the MCP server in Claude Code and Cursor when it finds them. The dialog keeps a copyable key as the fallback for a terminal on another machine.
 
-The live view is interactive: arrow keys walk the items, Enter opens actions (detail, apply this item with Claude Code, copy as prompt, mark applied / accepted / rejected / reopen, delete), `a` hands every open item to Claude Code in the repo, and the header shows who is reviewing right now in their own colours. Everything is the shared room state, so a status flipped in the terminal or by an agent changes in reviewers' browsers instantly, and vice versa.
+The live view is interactive: arrow keys walk the items, Enter opens actions (detail, apply with Claude Code, copy as prompt, mark applied / accepted / rejected / reopen, delete), `space` selects items and `a` hands them (or all open ones) to Claude Code or Cursor — you pick agent and model per task (Fable / Opus / Sonnet / Haiku…) and watch it work in an agent pane right under the list while statuses flip live; `f` filters, `/` searches — and the header shows who is reviewing right now in their own colours. Everything is the shared room state, so a status flipped in the terminal or by an agent changes in reviewers' browsers instantly, and vice versa.
+
+One repo, several environments (localhost, dev domain, production): pairing from each adds its origin to the project, the live view shows them all with a `where` column, `o` switches between them, `reviewkit env` manages the list.
 
 Everyday commands: `reviewkit list --status open`, `reviewkit show <id>`, `reviewkit status <id> applied`, `reviewkit watch`, `reviewkit export -o review.json`, `reviewkit prompt` (agent-ready work list), `reviewkit rooms`, `reviewkit projects` / `-p <name>` to act on another paired site. Full table in [`cli/README.md`](./cli/README.md).
 

package/dist/{chunk-EOX3UJNP.js → chunk-L7JLBD7Q.js}

@@ -33,4 +33,4 @@ function LazyReviewKit(props) {
 export {
   LazyReviewKit
 };
-//# sourceMappingURL=chunk-EOX3UJNP.js.map
+//# sourceMappingURL=chunk-L7JLBD7Q.js.map

package/dist/{chunk-5MAFDRUI.js → chunk-VNRVRNZX.js}

@@ -3750,6 +3750,11 @@ function currentAuthor(name) {
 
 // src/client/sync.ts
 var DEFAULT_SYNC_URL = "wss://2-59-219-26.sslip.io/rk-sync";
+function normalizeRoute(route) {
+  let r = String(route ?? "/");
+  if (!r.startsWith("/")) r = "/" + r;
+  return r.length > 1 ? r.replace(/\/+$/, "") : r;
+}
 function resolveSyncUrl(config) {
   return typeof config.syncUrl === "string" && config.syncUrl !== "" ? config.syncUrl : DEFAULT_SYNC_URL;
 }
@@ -3832,8 +3837,11 @@ function useRealtimeSync(state, rawDispatch, config, reducer2) {
           for (const id2 of deletedIds) dispatch({ type: "REMOTE_DELETE", id: id2 });
           dispatch({ type: "PEERS_SET", peers: peers.filter((peer) => peer.id !== id).map(toPeerState) });
           const serverIds = new Set(items.map((item) => item.id));
+          const here = normalizeRoute(window.location.pathname);
           for (const item of stateRef.current.session?.items ?? []) {
-            if (!serverIds.has(item.id) && !deleted.has(item.id)) sync.addItem(item);
+            if (serverIds.has(item.id) || deleted.has(item.id)) continue;
+            if (normalizeRoute(item.route) !== here) continue;
+            sync.addItem(item);
           }
         },
         onPeerJoin: (peer) => dispatch({
@@ -7660,4 +7668,4 @@ export {
   importSessionFile,
   ReviewKit
 };
-//# sourceMappingURL=chunk-5MAFDRUI.js.map
+//# sourceMappingURL=chunk-VNRVRNZX.js.map