Skip to content

Releases: stroops/sloop

v0.1.3

Choose a tag to compare

@github-actions github-actions released this 03 Jul 00:52
65c69e2

Every builtin AI CLI now reports precise fleet status, the status bar gained
model/context/rate-limit awareness, and sloop new launches an agent without
stealing your terminal.

Copilot and Codex auto-install status hooks

  • sloop hooks install now wires up all five hook-capable tools —
    claude, gemini, cursor, copilot, and codex — instead of stopping at
    print+paste for the last two.
  • Copilot: sloop owns a dedicated ~/.copilot/hooks/sloop.json (Copilot
    loads every *.json file in that directory), so install/uninstall never
    touches a hand-written hook file.
  • Codex: sloop claims its single notify slot in ~/.codex/config.toml
    only when it's free; if something else already runs there, hooks install
    prints copy-paste chaining instructions instead of overwriting it. A new
    hidden sloop hooks notify <tool> command routes Codex's one-program
    payload to the same status marker hooks emit already writes.
  • The manifest's per-state event mapping (hooks.events.*) generalized from
    a plain string to {event, matcher}, so a tool's event can be
    discriminated by a sub-type (Copilot's notification fires for several —
    only permission_prompt means "waiting").
  • Surveyed gemini-cli, Cursor CLI, and Copilot CLI for a native statusline
    mechanism (separate from hooks): none of the three expose one today, so
    that gap is now a documented, dated fact in each manifest rather than an
    open question.

Status bar: model, context %, rate limit — without repeating a tool's own footer

  • sloop statusline install <tool> (offered automatically by sloop init)
    registers a sloop feed on Claude/Antigravity's own statusline mechanism,
    chaining to any command you already had, and enriches the fleet marker
    with what the tool reports on every render.
  • The tmux bar and sloop ps show model, context-window usage (as a
    block-character bar), and git branch at a glance — moved to the left
    side, since tmux truncates the right side first on a narrow terminal.
  • New: 5-hour rate-limit usage, sourced from each provider's own convention
    — something no custom statusline script commonly surfaces, so it's always
    shown.
  • Once a tool's own footer already reports model/context/branch (a feed is
    wired), the tmux bar stops repeating them and narrows to what only sloop
    knows (status, identity, rate limit); tools with no feed (codex/cursor)
    keep the full picture.
  • One marker read and one pane capture now serve a whole render (previously
    up to three round trips); ~/.sloop/state no longer grows forever —
    sloop kill removes a session's marker immediately, and sloop ps prunes
    stale ones after 24h.

sloop new — launch an agent without attaching

  • new is run minus the attach: same targets/flags/context sync, but your
    terminal stays free (-a attaches anyway, -N forces a fresh instance).
    run and new share one launch path so the two commands can't drift.
  • Fixed a tmux target-matching bug: with ws__claude__sec running, a bare
    -t ws__claude could silently resolve to it instead of failing, so
    attach/send/kill risked hitting the wrong agent. Whole-session targets now
    always match exactly.

Also

  • Second-account profiles (config_dir_env, e.g. CLAUDE_CONFIG_DIR) are
    now respected everywhere a hook or statusline config path is resolved, not
    just the default account dir.

v0.1.2

Choose a tag to compare

@github-actions github-actions released this 29 Jun 17:39
4ba7a24

Changelog

  • 4ba7a24: feat: Home menu, sloop update, and consistent interactive screens (#10) (@tuanngo)
  • 4df8d0f: v0.1.2: one-key reach to waiting agents + second Claude account (#9) (@tuanngo)

v0.1.1

Choose a tag to compare

@github-actions github-actions released this 28 Jun 15:03
1205060

Run more than one agent per repo, peek into a waiting agent without leaving your
screen, and a sharper first-run experience — all on top of v0.1.0.

Run multiple agents & accounts

  • Named instances: sloop run claude@review / -n/--name runs a second agent of the same tool in one
    repo (session <repo>__tool__instance); -N/--new auto-names the next free slot (claude·2…).
  • Profiles: save a tool + env once with sloop profile add|ls|rm (global ~/.sloop/config.yaml) and
    launch it as sloop run @<name> — e.g. a different account via CLAUDE_CONFIG_DIR. --env KEY=VAL
    injects env one-off without a profile (~/$VAR expanded). The fleet view shows instances as
    tool·instance.

Peek (overlay a waiting agent)

  • sloop peek floats a waiting agent's live pane over your current screen so you can answer it and
    drop back without switch-client swapping your whole screen; sloop peek setup binds a key. Needs
    tmux ≥ 3.2. Every status bar gains a fleet-wide ⏳ N waiting badge.

Sharper onboarding (init / check / doctor)

  • sloop init is provider-respecting: it asks about the tools you actually use (Claude-first order),
    prompts honestly and skips work that is already done, selects tools per workspace, and no longer
    offers to scaffold a provider's folders for it.
  • sloop check gains more AI-readiness criteria, sourced from each tool's adapter manifest.
  • sloop doctor groups and colors its output and explains the mode line.

Full Changelog: v0.1.0...v0.1.1

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 28 Jun 07:15
b79917b

⚓ Sloop v0.1.0 — first release

The local-first control layer for your AI coding CLIs. One canonical context for every tool, and one cross-repo view of every agent you're running — sitting above Claude Code, Cursor, Codex, Copilot, Gemini & Antigravity, never replacing them or proxying their models.

Highlights

🧭 One canonical context, every tool
Write project guidance once in AGENTS.md; sloop delivers it to each CLI — pointer files (CLAUDE.md, GEMINI.md, …) created only if missing, never clobbering your files. Skills are authored once and symlinked into every tool.

🛰️ Cross-repo fleet view
sloop ps shows every running agent across all your repos and floats the ones waiting on you to the top. It reads what each agent is asking and lets you answer in one keystroke — jump in, reply, approve, or kill in place. Plus sloop ls (your workspaces), sloop restore (relaunch the whole fleet after a reboot), a floating sloop hud popup, and a live per-session status bar.

🔌 Provider-aware by construction
Every tool is a declarative adapter manifest — adding a CLI is adding one file, never editing Go. Status hooks wire each tool's own lifecycle events, so the fleet view knows precisely who's waiting (Claude, Gemini & Cursor auto-install).

🪶 Local-first & lightweight
A single CGO-free Go binary. No daemon, no cloud, no bundled LLM. tmux on macOS/Linux, psmux on native Windows.

Install

brew install stroops/tap/sloop
# or
go install github.com/stroops/sloop/cmd/sloop@v0.1.0

Then sloop init in a repo, write your AGENTS.md, and sloop run claude.

Full notes in the CHANGELOG · guide in the README.