v2.0.1

Changelog

• 00d3dd1: Release 2.0.1: the cask stops stripping the quarantine (Claude Opus 5.5 noreply@anthropic.com)

v2.0.1-rc.1

Changelog

• fd8f8a2: Accept both Go versions' wording of a JSON type error (Claude Opus 5.5 noreply@anthropic.com)
• 1f4b75c: Keep the output of a process that exits at once (Claude Opus 5.5 noreply@anthropic.com)
• adfd028: Move to pnpm 12 (Claude Opus 5.5 noreply@anthropic.com)
• 8b4d4cf: Pin actions to commit SHAs (Claude Opus 5.5 noreply@anthropic.com)
• 277c990: Sign and notarize the darwin binaries (Claude Opus 5.5 noreply@anthropic.com)

v2.0.0

Changelog

• 8cd4637: Release the tag that triggered the run (@kondaurovDev)

v2.0.0-rc.1

Changelog

• 8e3f0d3: Install channels beyond npm: Homebrew cask, install.sh, deb and rpm (@kondaurovDev)
• 245e69f: Merge origin/main (1.1.1 release notes) into go (@kondaurovDev)
• ed17da7: Move the UI to Effect 4.0 stable and foldkit 0.168 (@kondaurovDev)
• c1a0400: Observer forwards x-forwarded-* and propagates aborts (@kondaurovDev)
• 965364b: Port procdeck to Go: one binary with the UI embedded (@kondaurovDev)
• 5bea0dc: Publish to npm through trusted publishing, no token (@kondaurovDev)
• a38f937: Read env files natively, follow proc logs, rotate the deck log (@kondaurovDev)
• 32daac9: Release pipeline for 2.0: GoReleaser, npm launcher, release candidates (@kondaurovDev)
• 234c4c6: chore: release packages (@github-actions[bot])

procdeck@1.1.1

Patch Changes

• c1a0400: HTTP observer forwards standard proxy headers and propagates connection failures faithfully:

  • x-forwarded-host now carries the original Host (an outer proxy's value wins), so frameworks that CSRF-check Origin against the forwarded host — Next.js Server Actions — no longer reject mutating requests. x-forwarded-proto and x-forwarded-for are set too.
  • A client abort (closed tab, cancelled fetch) now destroys the upstream request instead of letting it stream to completion unseen.
  • An upstream dying mid-body truncates the client response by cutting the connection, instead of hanging the client or corrupting the payload with an appended error note.

procdeck@1.1.0

Minor Changes

• dace5bb: A proc can pin its assigned ${port} to a fixed public number: port: 8787
  on the spec. The public side stays exactly that number — dotenv files, mobile
  simulators and teammates' scripts keep their hardcoded localhost:8787 —
  while an observed proc still binds a hidden internal port behind the HTTP
  observer, so the traffic gets captured with no wiring changes anywhere.
  Conflicting pins (duplicate, on the UI's port, or on a proc that never uses
  ${port}) are config errors; a pin whose port is already taken parks just
  that proc in blocked — naming the port and the likely holder — while the
  rest of the deck runs, and Start retries the bind. A pinned proc with no
  url also gets its UI link derived from the pin.

procdeck@1.0.0

Major Changes

• 2364ad0: procdeck 1.0.0. Nothing breaks in this release — the version says the surface
  is settled and will not move under you without a major bump.

  What 1.0 promises to keep stable:

  • The config, both formats: every field of the schema (and therefore
    schema.json), the ${port} / ${port:id} templating, needs /
    readyWhen readiness, preflight, alerts, observe, and the
    defineConfig export.
  • The CLI: the verbs (up, down, restart, status, ls, open,
    init, logs, mark, wait-for, errors, http, agents, mcp),
    their flags, and the shape of every --json payload. wait-for keeps exit
    2 for a timeout.
  • The MCP tools: names and parameters of deck_status, get_logs,
    get_http, get_errors, set_mark, wait_for, timeline, and the
    --mutations trio.
  • The addresses: <id>.localhost:<ui-port> for every proc, the UI on
    4820 by default, loopback-only unless host says otherwise.

  Internals stay internal: the deck's HTTP API, the registry files under
  ~/.procdeck/, the SSE event shapes and everything under src/ are free to
  change.

  The readme is now the pitch and the docs live in docs/ — config reference,
  the agent harness, the traffic observer and the architecture notes.

Minor Changes

• 6ef989d: procdeck restart --all restarts every deck on this machine — stop, then detach again
  from each deck's own registry entry (foreground decks are skipped). The version a deck
  actually runs now shows in status, ls and the UI's deck switcher, and status/up
  say so out loud when it differs from the CLI — the answer to "did my update reach this
  deck?".

procdeck@0.5.0

Minor Changes

• 10b557f: procdeck is now a harness for coding agents, not just a dashboard for humans.
  New CLI verbs, all bounded and --json-able: status --json grows an
  attention list (crashed / blocked / alerting procs with reasons — the
  cheapest first question); logs [proc] returns pane output as plain
  timestamped lines (ANSI stripped, \r progress bars keep only their final
  state), filterable with --grep, --since, --lines — no proc means every
  proc interleaved with [id] prefixes, and the old daemon-log behaviour moved
  to logs --self [-f]; mark [name] drops a named marker in every proc's
  stream and --since-mark shows only what happened after it — the verify
  loop (mark → edit → restart → check) in four commands; wait-for <proc> [--pattern RE] blocks until a listening port or a matching line, failing
  fast with the log tail if the proc crashes instead (exit 2 on timeout);
  errors [proc] parses stack traces out of recent output and deduplicates
  them by signature ("same TypeError, 41×, last 3s ago"); restart <proc>
  restarts a single process. procdeck mcp serves the same verbs over MCP
  (stdio) — claude mcp add procdeck -- procdeck mcp once, globally, and the
  instance registry finds the right deck per project; read-only by default,
  --mutations adds restart/stop/start, plus since_last cursors and a
  timeline tool. procdeck agents (and procdeck init) plant the discovery:
  a ## procdeck section in CLAUDE.md / AGENTS.md and a Claude Code skill
  teaching the loop.
• 10b557f: procdeck sees the traffic now, not just the logs. Procs that use ${port}
  get an HTTP observer interposed on their assigned port: the proc binds a
  hidden internal port, procdeck listens on the public one and forwards — so
  requests between processes (${port:api}) and into them are captured
  transparently, statuses and bodies included, with zero app changes
  ("observe": false opts a proc out; the *.localhost proxy captures for
  opted-out procs). WebSocket connections are captured per-message with
  direction and text — compression is stripped at the handshake so frames stay
  readable. Bodies are text-only and truncated (16 KB); authorization,
  cookie and friends are redacted before anything is stored. Three surfaces
  over the same per-proc rings: procdeck http [proc] [--status 5xx|422|error] [--path RE] [--since-mark] [--ws] [--body] [--digest] [--json] — --digest
  groups 4xx/5xx by route with path params collapsed (/users/:id); a
  get_http MCP tool with since_last cursors (and timeline now carries
  the window's exchanges); and a traffic view in the UI — the ⇄ position of
  the layout switch — with kind/proc/errors filters, click-to-expand bodies,
  pause and clear. Marks span both streams, so mark → act →
  http --since-mark shows exactly which requests a change caused and what
  they returned.

procdeck@0.4.0

Minor Changes

• 8e4b627: procdeck runs in the background now. procdeck up (and bare procdeck) starts the
  deck detached, opens the UI and returns — no terminal tab kept hostage; up --fg is
  the old foreground mode. New commands, all working from anywhere inside the project:
  down (terminates every process tree), restart, status, ls (every running deck
  on the machine), open and logs [-f]. Running decks register in
  ~/.procdeck/instances/ (stale entries pruned by pid; PROCDECK_HOME relocates it);
  a detached deck's own output goes to ~/.procdeck/logs/. up refuses a port already
  held by another deck — naming it — or simply busy, before anything spawns; a bad
  config is reported in the terminal, not in a log file. In the UI: ⏻ shuts the deck
  down (with a confirm), the page says so instead of "reconnecting" and revives itself
  when procdeck up runs again; the deck name in the bar lists the other running decks
  as links. The CLI moved to effect/unstable/cli — --help per command, shell
  completions via --completions.
• f2abc61: Three things for the first real release. Loopback only: the server binds
  127.0.0.1 by default — the UI types into real terminals, so it is not something to
  put on the LAN by accident; "host": "0.0.0.0" in the config opens it up for
  devcontainers and VMs. Per-proc backlog: each proc keeps its last 256 KB of output
  and a new tab gets every pane's history plus a synced marker before live events — a
  chatty ticker can no longer evict a quiet server's startup lines, a deck that has run
  for days still opens with every pane populated, the UI knows exactly which chunks are
  news, and a reconnect replays into freshly reset panes instead of appending history to
  itself. procdeck init: writes a first procdeck.config.json from what is already there —
  a Procfile, workspace packages with dev scripts (via the package manager your lockfile
  points at), plain backend/ + frontend/ subdirectories (each its own package.json or
  a Django / Go / Rust / Rails / compose project, with cwd set), or the root itself.

Patch Changes

• 9f944e5: The deck tells you when it needs you. Procs that crashed, are blocked on a preflight,
  carry an alert or have unread error lines count into the tab title ((2) garage · procdeck) and put a red dot on the favicon — visible from the tab strip and the Dock.
  A bell in the global bar turns on system notifications: a crash, a block or a new alert
  while the tab is hidden or the window unfocused shows one (click brings the deck back);
  replayed history on reconnect never notifies. The bell state persists with the UI state;
  the browser permission is asked for on the first click.

• 1db7f78: Grid pinning. Pin a proc (📌 in its pane header or sidebar row, ⌥P on the active one) and
  the grid shows only pinned procs; the rest collapse into a tray strip under the tiles —
  status dot, id, badges — where a click peeks at one and the pin puts it back. Nothing
  pinned = everyone tiled, as before. ⌥Z zooms the active tile in and out of single. The
  grid now sizes its columns from the tile count (near-square, at most 4) so the deck fits
  the screen when it can and scrolls only when tiles would drop under 180 px. Pins persist
  with the rest of the UI state.

• e90a08c: Polish before publishing, part one.

  • URLs in pane output are clickable (Local: http://localhost:5173/ from every dev
    server opens in a new tab).
  • Exit status reads like a human wrote it: exit 1 · 2m ago, killed (SIGTERM) · 5s ago
    — signal names instead of numbers, and no more signal 0 on a plain non-zero exit.
    Running procs that were respawned show a restart count (↻3), so crash loops are
    visible from the status line. The sidebar shows the same exit summary where uptime sits
    while running.
  • Grid tile headers keep the *.localhost address whole instead of clipping it to :61:
    compact tiles show only the primary address (raw ports in its tooltip) and the status
    text is what gives way first.
  • The assigned port no longer also counts as "internal" when it lands in the ephemeral
    range.

• a85d537: Light theme. A System · Light · Dark switch in the global bar; System follows the OS
  (prefers-color-scheme) and tracks it live. The terminals switch palette too — light gets
  its own ANSI set (GitHub Light), since xterm's defaults are unreadable on white. The choice
  is stored with the rest of the UI state and painted before the first frame, so there is
  no flash on reload.

• 512e814: The UI remembers itself across reloads, and says when it has lost the server.

  Layout (single/grid) and the selected pane are kept in localStorage — per origin, so per
  deck — and restored on the next load; a stored pane that is no longer in the config falls
  back to the first one. While the event stream is down (a procdeck restart, a killed
  server) a "reconnecting" banner sits over the panes instead of the UI silently showing
  stale state; when the stream is back the proc snapshot is refetched so statuses that
  changed in the meantime catch up. A stream the browser gave up on (server answered with
  something other than an event stream) is retried every 2 s.

procdeck@0.3.0

Minor Changes

• a1ced2e: Cut the install from 116 MB to 1.8 MB, and remove every reason for it to fail.

  node-pty is replaced by @lydell/node-pty — the same upstream sources and the same API,
  but the binaries ship as per-platform optional dependencies instead of one tarball carrying
  every platform's prebuilds. Upstream runs an install script and has no Linux prebuild at
  all, so Linux users needed a C++ toolchain and pnpm ≥ 10 / Bun users needed
  pnpm approve-builds (or trustedDependencies) on every platform or spawning failed at
  runtime. The replacement runs no install script and compiles nothing, anywhere — which also
  retires the startup hack that repaired the darwin spawn-helper execute bit.

  effect is no longer installed either: it is bundled into the CLI, where tree-shaking takes
  it from 48 MB on disk to under 1 MB of shipped code. The package's public entry point
  (defineConfig and the config types) is now a standalone module with no effect types in its
  declarations, so a procdeck.config.ts typechecks against a package that has none of
  procdeck's own dependencies installed.