/statusline — Magician CLI status line

Bundled command: if magician-ui is not found, run it as ${CLAUDE_PLUGIN_ROOT}/tools/magician-ui. If that is missing too (Claude chat ships no plugin tools), say the status line needs Claude Code.

A native Claude Code status line rendered by magician. It runs locally, consumes zero API tokens, updates on each message (debounced), and helps you catch context rot before it bites. It's driven by the bundled magician-ui CLI, which edits ~/.claude/settings.json safely (timestamped backup → validate → atomic write; it never leaves settings broken). Backups are mode 0600 and only the newest 3 are kept; an edit that changes nothing writes nothing. enable, set and disable touch only the statusLine key; cleanup removes only entries on the fixed list older magician versions used, and only when cli-ui.json records that they added them (a matching rule the user added too goes with them; see below). Enabling copies the renderer to ~/.claude/magician/statusline.py, which the statusLine command points at.

The status line is off until the user asks for it: magician never enables it on its own. Every command that changes settings (enable, set, disable, cleanup, voice <level>, ste on|off) goes through the normal permission prompt; only the read-only status commands are pre-approved.

When to use

Use to enable, configure or disable the status line (context %, model, git, cost, skill, effort, voice), set the output voice, turn STE-lite (simplified technical English) on or off, or clean up the permission rules older versions left. Trigger phrases: "enable/turn on the magician UI / status line / status bar", "show my context/tokens/effort in the console", "configure/change what the bar shows", "turn off / disable the status line", "make output shorter/leaner / set the voice", "make output STE-lite / simplified technical English", "clean up magician's old permission rules". Not for general terminal or theme questions, and never enabled on the skill's own initiative.

Components (user-configurable)

Actions

• Enable — only when the user asks. First confirm which components they want (default: all) with AskUserQuestion (multi-select), unless they already said (e.g. "just rot and context"). Then:

  magician-ui enable --all              # everything
  magician-ui enable --only context,rot # a chosen subset

  A plain magician-ui enable (for example to refresh the renderer after an update) keeps the saved components.
• Change what shows (no re-enable needed):

  magician-ui set context,rot,spark

• Status — magician-ui status (state, components, whether it's wired into settings, and any permission entries an older magician version left behind).
• Disable — magician-ui disable removes only magician's statusLine entry; re-enable anytime on request. magician-ui disable --purge also deletes magician's status-line files (the copied statusline.py, the status/ markers and cli-ui.json). It refuses, changing nothing, while cli-ui.json still records entries for cleanup to remove; disable --purge --force deletes the record anyway and leaves those entries in settings. It also refuses while cli-ui.json can't be read and settings still has entries an earlier version could have added (allow rules on magician's old list, or defaultMode auto), since the record might cover them. Suggest /statusline disable before the user uninstalls magician, so statusLine doesn't keep pointing at the copied renderer.
• Clean up after older versions — magician 4.14 and earlier could add permission rules, permissions.defaultMode and an auto-mode env key to ~/.claude/settings.json. magician-ui cleanup removes those entries only where cli-ui.json records that magician added them (the allow rules, and defaultMode set to auto), plus the env key and obsolete magician data files. Matching allow rules with no record stay, and cleanup names them; a rule the user also added while the record exists can't be told apart and is removed with magician's, except Bash(jira:*) and Bash(confluence:*) when the record is from 4.4 or later. Run cleanup before disable --purge, which deletes the record (and refuses to while the record is still needed, unless --force is given). magician-ui status lists what it would remove. Run it only when the user asks, and tell them what it will change first.
• Unreadable cli-ui.json — if the file exists but can't be read (not valid JSON, or not a JSON object), magician never replaces it. enable, set, default, lore on|off, voice <level> and ste on|off stop and change nothing; cleanup and disable do the rest, leave the file as is and say so; status, lore status, voice status and ste status note it. Session start keeps reading the file's old values (status markers, the weekly upgrade notice) until it's fixed or deleted. Suggest the user fix the file (then run cleanup again if it recorded old entries) or delete it.
• Permission mode — magician does not change Claude Code's permission mode or rules. To start in auto mode, the user can press Shift+Tab, pass --permission-mode auto, or use /config; to save rules, /permissions.
• Voice — output brevity (lower token cost, no quality loss) — magician-ui voice warrior|scribe|bard sets how wordy responses are. Output tokens cost several× input, so leaner output is the cheapest saving. Levels least→most wordy: warrior (minimal but complete), scribe (the default — leaner than usual), bard (standard/native). For warrior/scribe, SessionStart adds a brevity style note to the session context that cuts filler (preambles, recaps, restating the request) while keeping all substance and code/commands/errors verbatim — it never compresses prose into fragments or jargon. magician-ui voice status shows the current level. Overrides (first match wins): env MAGICIAN_VOICE → per-project .magician/voice → this setting → default scribe. Takes effect next session start; the 🗣 voice: chip shows it live.
• STE-lite — simplified technical English (grammar, not length) — magician-ui ste on|off turns the STE-lite prose rules on or off. The voice sets how much Claude writes. STE-lite sets how each sentence is built: 25 words or fewer in a descriptive sentence, 20 or fewer in a step, one instruction per step, imperative steps, the condition before the command, active voice, simple tenses, no semicolons, no contractions, no Latin abbreviations. Code, identifiers, numbers and error text stay verbatim. The rules are STE-inspired and paraphrased. They make no claim of ASD-STE100 compliance. Default off. When on, SessionStart adds one short STE note after the voice note. magician-ui ste status shows the state. Overrides (first match wins): env MAGICIAN_STE=1|0 → per-project .magician/ste (content on or off) → this setting → off. Takes effect next session start. For one session only, /output-style magician:ste switches the response style without the flag. The full rule layer is in lore/ste.md.

Rules

• Act only on the user's request. Magician never enables the bar, re-adds it, or suggests it on its own; respect a "no".
• Enabling/disabling changes ~/.claude/settings.json; it hot-reloads, so the bar appears/updates within a message or two — no restart needed.
• Keep it lightweight: recommend a smaller component set if the user wants a minimal bar (e.g. context,rot).
• Never hand-edit settings.json for this — always go through magician-ui, which backs up + validates.

Obstacles

This is a configuration utility that reports to the human directly, not upward to an orchestrator — so it emits no upward Obstacles block: surface any blocker in prose to the user and stop.

See lore/obstacles.md.

Completion Signal

"Magician CLI UI <enabled (components: …) | updated | disabled | cleaned up>. It runs locally (no tokens) and hot-reloads."