Skip to content

CHANGELOG.md

Latest commit

 

History

History
277 lines (213 loc) · 12.4 KB

CHANGELOG.md

File metadata and controls

277 lines (213 loc) · 12.4 KB
Raw

Changelog

All notable changes to workspace are documented here. The format follows Keep a Changelog; versions follow Semantic Versioning.

Unreleased

(Changes land here as they merge; a release retitles the section. The installed daily driver may be ahead of the last tag — workspace --version says exactly how far when built with git describe.)

Added

  • The skill teaches coordinating with sibling sessions via Claude Code's native cross-session messaging: before reworking or removing a shared interface (and after landing a breaking change), a session warns the sessions whose workspaces have that project checked out (ls -g names them; a bare directory under the root is not a reason); an incoming warning is weighed as advisory peer input. Live-only by design — durable cross-session facts stay in the workspace's ## Status note. (The durable announcements board is a separate, future feature.)

1.7.1 — 2026-08-24

Changed

  • README rewritten as a front door (~400 lines, short sentences, primary info only); the deep contracts moved to docs/reference.md, linked per-section. The v1-divergences appendix is gone — readers have no v1; every still-relevant behavior is stated as a plain fact where it belongs. No behavioral claims were added or dropped (independently audited).

1.7.0 — 2026-08-21

Added

  • Onboarding: the Claude Code skill teaches sessions the configure-my- project procedure (read the repo, draft the config entry, route every shared resource through ${} values, change the app to consume the env, prove it with a disposable workspace) and triggers on "i installed it, now what" / "configure workspaces for my project". A worked Rails + Vite

    • Sidekiq example — config entry plus the app-side checklist — lives in examples/rails.md, linked from the README.

  • The skill teaches handoff reports: a session's last message is what the user sees when switching back to a workspace, so it ends with Done / Needs you / Watch out — asks self-contained and decidable, journeys compressed, decisions never omitted.

  • The status note: sessions maintain a ## Status section in the workspace's CLAUDE.md (About / Now / Next / Needs, with an as-of date — the handoff report made durable); workspace status <ws> renders it verbatim, so "should I get back to this?" is answerable from outside any session — and the session-start hook delivers it to every session that reopens the workspace. The tool records nothing: it displays a file sessions own, pids-dir style.

Changed

  • The listing sheds its duplicate columns. ls (and no-argument status) show WORKSPACE INDEX NOTE: the name IS <task>_<description-slug>, so the TASK and DESCRIPTION columns printed it twice more and tripled the table's width (worse for adopted workspaces, whose "task id" is the whole dir name). Names clip at 60 runes with a visible …; the NOTE cell carries (adopted) and the -a labels; the real description stays in status <ws> and --json (all fields unchanged). The skill now also tells sessions that a workspace description is a NAME, not a summary — a few words, not a sentence.

Fixed

  • Completion matches case-insensitively: cd try<TAB> offers the TRY-* workspaces. The completed word always carries the candidate's real casing.

1.6.0 — 2026-08-18

Added

  • workspace try <description…> — one command to a thinking room: a projectless workspace with a generated TRY-<n> id, the description taken from every word you type (no quotes), and a session opened in it. Allocated values included; graduate with checkout, discard with destroy. The id generator counts released drafts too, so numbers are never reused against a surviving directory.

1.5.0 — 2026-08-14

Added

  • ls -a/--all: also lists root directories no allocation claims — released workspaces (the tool's .workspace footprint identifies them; task id derived from the dir name; adopt to reuse) and stranger dirs, labeled (unmanaged). The registry stays allocations-only: "archived" is a condition of the world, derived on demand, never a recorded status.
  • doctor notes every unregistered dir in the root (uncounted) — the always-on discoverability net for the same state.

1.4.0 — 2026-08-14

Changed

  • Worktree branches are named after the full workspace name (PATED_patternima-editor-fixes), not the bare task id — the branch now says in git branch and in a PR list what it is about. Applies to new checkouts; existing workspaces keep their task-id branches and every command keeps working with them (branches are always read from the worktree, never recomputed).
  • gc --destroy-dirs's merged gate now checks each project's actual branch instead of recomputing it from the task id — which also makes the gate honest for adopted workspaces and hand-switched worktrees, and reads an unreadable branch as "keep".

1.3.6 — 2026-08-14

Added

  • workspace ls (and the no-argument status listing) print column headers: WORKSPACE INDEX TASK DESCRIPTION, plus PROJECTS under -g. The empty listing keeps its plain "no workspaces" line. --json is unchanged.

1.3.5 — 2026-08-14

Changed

  • browse dials the port before opening: nothing listening is a refusal that names the workspace up to run (and hands over the URL), instead of a browser tab pointing at a dead port. The socket is asked directly — a hand-started server counts, daemon records are not consulted. The skill teaches the sequence: start what serves, then browse or verify.

1.3.4 — 2026-08-14

Added

  • doctor reports a finding for a browse_port whose ${…} template does not resolve to a port number — the broken-config case load validation cannot see (token resolution needs the values math). Complements 1.3.3: the typo class is now caught at load, at doctor, and at browse.

1.3.3 — 2026-08-14

Fixed

  • browse no longer opens impossible URLs. A browse_port that is neither a number nor a ${VALUE} template is now a config error at load with a did-you-mean hint (browse_port: PORT0 → did you mean "${PORT0}"? — the observed real-use typo opened http://localhost:PORT0); a template whose token fails to resolve is a loud error at browse time instead of a browser tab full of garbage.

1.3.2 — 2026-08-14

Changed

  • Workspace completion offers full names only. Task ids still resolve when typed, but completing both listed every workspace twice — and an id is always a prefix of its name, so a typed id-prefix reaches the full name (which, unlike a bare task number, says what the workspace is about).

1.3.1 — 2026-08-12

Changed

  • Docs/history hygiene: internal host names scrubbed from the repository and its history (module-proxy consumers should use this version or later). README gains the "Separation, not virtualization" section and the changelog itself. No code changes.

1.3.0 — 2026-08-12

Added

  • Live progress for the ensure chain. new, checkout, up and launch now report each slow step as it runs — the worktree checkout and every setup: command — one line each, completed in place with its duration ( app: setup: npm ci… ok (11.6s)). Idempotent re-runs stay as quiet as before: only work that actually runs reports. Plain text, no spinners, pipe-safe; setup output itself stays captured.
  • launch tips the parallel terminal. On the create path, launch now prints tip: in another terminal: workspace cd <id> — work alongside this session instead of the hint: workspace cd line, which read as a pending to-do right before Claude took over the terminal. new keeps the hint (there it is the next step); launch's reuse path prints neither.

1.2.0 — 2026-08-12

Added

  • Automatic session titles. claude and launch sessions title the terminal (OSC escape, when stdout is a terminal, first 40 characters of the workspace name) and rename the current tmux window (tmux rename-window, first 20, when inside tmux). When the session ends, the window's automatic-rename option is unset — the exact undo of the rename — so tmux auto-naming resumes on default configs and manual-name configs stay untouched. Best-effort and silent: no tmux, no tty, no problem. (The return of v1's title command, automated — the first dropped command daily use proved missed.)
  • MIT license; GitHub Actions CI (full suite + race detector on Linux, build + vet on macOS); README badges, go install instructions and a "How this was built" section.

1.1.0 — 2026-08-11

Changed

  • Daemons are lazy. launch no longer starts daemons (and no longer restarts dead ones when reusing a workspace). Start what you need with workspace up <ws> [target…] — the session-start status block tells the session what exists. Breaking relative to 1.0 behavior.

Added

  • Per-daemon description:. start: entries accept a third shape, {name: {command, description}}. Descriptions render — with ${} values substituted — in workspace status (human and JSON), in WORKSPACE.md's new per-project services list, and reach every Claude session through the SessionStart hook, so a session knows what each daemon is for before starting it.
  • workspace doctor notes (uncounted, never a finding) every configured daemon without a description.
  • The skill and the SessionStart hook teach the lazy-daemons convention.

1.0.2 — 2026-08-11

Changed

  • Module path renamed to github.com/Phaengris/claude-workspaces (the public home); the installed skill directory sheds its -go suffix.

1.0.1 — 2026-08-11

Fixed

  • Shell completion for workspace launch: project names now complete after the description slot (and workspace identifiers in the first slot when flags precede it).

1.0.0 — 2026-08-10

Initial release: the complete environment engine, a clean-room Go rewrite of an earlier personal Ruby tool (never publicly released).

  • Workspaces: new, checkout, destroy, adopt, release, gc — a git worktree per project, an allocated index deriving ports and other values, generated .env, WORKSPACE.md/CLAUDE.md scaffolding. Derived state throughout: the only registry is the allocations file.
  • Services: up, down, restart, logs with project:daemon targets, pid/starttime liveness, TERM→KILL escalation, run-and-wait preludes and stop: epilogues.
  • Sessions: claude and launch with flag injection, history-aware --continue, -- passthrough, curated-vs-inherited two-tier environment.
  • Introspection: ls, status, ports, env, which, cd, browse, doctor; meaningful exit codes (0/1/2/3/4).
  • Install: install/uninstall (manifest-driven), embedded Claude Code skill, SessionStart hook, shell wrappers and completions for fish/bash/zsh.