claude-token-saver 2.5.0 → 2.7.0

package/README.md

@@ -12,7 +12,7 @@
 **Why I built this.** I'm on the Max plan. On Opus 4.6 I never hit the *current-session* cap. After Opus 4.7 rolled out, I started hitting it on the same workflow — repeatedly. The official token statistics didn't match what I was actually feeling, and Claude Code's UI doesn't show prompt-cache health. This tool is what let me see *why*: low cache hit rate, 5m TTL writes that should have been 1h, 1M context auto-promoted in the background.
 
 v2.1 (2026-04) adds the workflow that follows the diagnosis:
-- **`claude-token-saver install`** — one command writes a Claude Code Skill (auto-activates when you mention "cache hit rate" / "1M context" / etc.) and a `/token-monitor` slash command.
+- **`claude-token-saver install`** — one command writes a Claude Code Skill that auto-activates when you mention "cache hit rate" / "1M context" / "5H cap" — no slash command needed.
 - **`claude-token-saver history`** — every warning chip transition is auto-logged to a daily Markdown file, so you can answer "when did this start" without grepping logs.
 - **Cross-platform paths** — Windows (`%APPDATA%`), macOS (`~/Library/Application Support`), Linux (`~/.config` / XDG) all handled.
 
@@ -208,11 +208,12 @@ One command wires up everything else this README mentions:
 claude-token-saver install
 ```
 
-This writes two files under your Claude user dir:
-- `~/.claude/skills/claude-token-saver/SKILL.md` — auto-activates whenever you mention chip wording ("⚠ 1M ON", "cache miss", etc.) or ask about token usage. Claude Code will then know to read history, drill into the table report, and explain the warning.
-- `~/.claude/commands/token-monitor.md` — adds a `/token-monitor` slash command that runs `claude-token-saver history` + a fresh report and summarizes both for you.
+This writes one file under your Claude user dir:
+- `~/.claude/skills/claude-token-saver/SKILL.md` — auto-activates whenever you mention chip wording ("⚠ 1M ON", "cache miss", "5H cap", etc.) or ask for a token report. Claude Code will then read `claude-token-saver last`, drill into history if needed, and explain the warning.
 
-Re-run with `--force` to overwrite. Install only one piece with `install --skill` or `install --command`.
+If you previously installed v2.5.x or earlier, `install` also removes the now-redundant legacy `~/.claude/commands/token-monitor.md` slash command — its workflow is fully absorbed into the skill (same behavior, triggered by intent rather than typing `/token-monitor`).
+
+Re-run with `--force` to overwrite the skill file.
 
 ## Warning history (`history`) — new in v2.1
 
@@ -286,7 +287,7 @@ That writes `./HANDOFF-YYYY-MM-DD-HHMM.md` in the current directory with:
   Read the most recent HANDOFF-*.md in this directory and continue the work.
   ```
 
-The handoff write is also recorded in history (`📝 handoff written: …`), so `/token-monitor` and `claude-token-saver history` show both the cap-warn and the backup event next to each other.
+The handoff write is also recorded in history (`📝 handoff written: …`), so `claude-token-saver history` and the auto-skill show both the cap-warn and the backup event next to each other.
 
 ## Hook Setup
 

package/bin/cli.js

@@ -201,9 +201,9 @@ function findLatestWarning(historyEntries, chipToCodes) {
 
 async function main() {
   // Subcommand: last — print the most recent warning + how to handle it.
-  // Designed for the /token-monitor slash command and the auto-skill so the
-  // user immediately sees "what just fired and how to fix it" without having
-  // to read the whole history file.
+  // Designed for the auto-trigger skill so the user immediately sees
+  // "what just fired and how to fix it" without having to read the whole
+  // history file.
   //   claude-token-saver last           # search last 1 day
   //   claude-token-saver last --days 7  # widen the lookback
   if (args[0] === 'last') {
@@ -329,38 +329,31 @@ async function main() {
     return;
   }
 
-  // Subcommand: install — write the Claude Code Skill and slash command so
-  // /token-monitor and the auto-trigger skill become available without any
-  // manual file editing. Cross-platform (uses node:path + node:fs).
-  //   claude-token-saver install              # install both
-  //   claude-token-saver install --skill      # only the skill
-  //   claude-token-saver install --command    # only the slash command
-  //   claude-token-saver install --force      # overwrite existing files
+  // Subcommand: install — write the Claude Code auto-trigger skill so the
+  // user can just mention chip wording and Claude responds. v2.6.0 dropped
+  // the redundant /token-monitor slash command in favor of the skill alone;
+  // a legacy command file is removed automatically. Cross-platform.
+  //   claude-token-saver install              # install/update the skill
+  //   claude-token-saver install --force      # overwrite existing skill file
   if (args[0] === 'install') {
-    const { installSkill, installCommand, installAll } = await import('../src/installer.js');
+    const { installAll } = await import('../src/installer.js');
     const force = hasFlag('--force');
-    const onlySkill = hasFlag('--skill');
-    const onlyCommand = hasFlag('--command');
     const print = (kind, r) => {
       const verb = r.action === 'exists' ? 'already exists' : r.action;
       console.log(`  ${kind}: ${r.path} (${verb})`);
     };
-    if (onlySkill && !onlyCommand) {
-      print('skill', installSkill({ force }));
-    } else if (onlyCommand && !onlySkill) {
-      print('command', installCommand({ force }));
-    } else {
-      const r = installAll({ force });
-      print('skill', r.skill);
-      print('command', r.command);
+    const r = installAll({ force });
+    print('skill', r.skill);
+    if (r.legacy.action === 'removed') {
+      print('legacy /token-monitor', r.legacy);
+      console.log('  (consolidated into the skill — same workflow, triggered by intent)');
     }
     console.log('');
-    console.log('Open Claude Code in any directory and try:');
-    console.log('  /token-monitor');
-    console.log('Or just mention "cache hit rate" / "1M context" — the skill auto-activates.');
+    console.log('Open Claude Code in any directory and just mention:');
+    console.log('  "cache hit rate" / "1M context" / "5H cap" — the skill auto-activates.');
     if (!force) {
       console.log('');
-      console.log('Tip: re-run with --force to overwrite existing files.');
+      console.log('Tip: re-run with --force to overwrite the existing skill file.');
     }
     return;
   }
@@ -562,8 +555,8 @@ async function main() {
   // refresh. Pull rate_limits + model out of it so we can surface cap-warn
   // (>=90%) chips, always-on usage segments, the model chip, record cap
   // transitions, and seed the table view's warning box. The table path falls
-  // back to the most-recent cached snapshot so the /token-monitor slash
-  // command (which doesn't pipe stdin) still has the data.
+  // back to the most-recent cached snapshot so the table view (which
+  // doesn't pipe stdin) still has the data.
   const stdinJson = readStdinJson();
   let caps = extractCaps(stdinJson);
   let model = extractModel(stdinJson);
@@ -616,7 +609,7 @@ async function main() {
       }
     }
     // Persist transitions to ~/.config/claude-token-saver/history/YYYY-MM-DD.md
-    // so /token-monitor and `claude-token-saver history` can replay them.
+    // so `claude-token-saver history` and the auto-skill can replay them.
     try {
       const { recordChip, recordCapTransition } = await import('../src/history.js');
       recordChip(spikeChip, { detail: chipDetail });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-token-saver",
-  "version": "2.5.0",
+  "version": "2.7.0",
   "description": "Save tokens on Claude Code — spike diagnosis, 1M-context detection, TTL countdown, statusline. (formerly claude-cache-monitor)",
   "type": "module",
   "bin": {

package/src/advice.js

@@ -59,6 +59,48 @@ export const ISSUE_MESSAGES = {
         commands: [`Press ${toggleShortcut()} to toggle on/off instantly`],
         commandsKo: [`${toggleShortcut()} 누르면 즉시 on/off 토글`],
       },
+      {
+        label: 'Cap extended-thinking budget (⚠ check /effort first)',
+        labelKo: '확장 사고(thinking) 예산 제한 (⚠ /effort 먼저 확인)',
+        commands: [
+          '⚠ Run `/effort` to check current level — `xhigh` is the #1 cap killer',
+          '/effort medium  — Anthropic-official default (use this for normal coding)',
+          '/effort low     — for simple edits, labeling, boilerplate',
+          '/effort xhigh   — ONLY for complex architecture / multi-file refactor planning, then revert',
+          'export MAX_THINKING_TOKENS=8000  # global hard cap (overrides /effort)',
+        ],
+        commandsKo: [
+          '⚠ `/effort`로 현재 단계 확인 — `xhigh`가 캡 소진의 1순위 원인',
+          '/effort medium  — Anthropic 공식 기본값 (일반 코딩은 이걸로)',
+          '/effort low     — 단순 편집·라벨링·보일러플레이트',
+          '/effort xhigh   — 복잡한 아키텍처·다파일 리팩터 계획 한정, 끝나면 즉시 복귀',
+          'export MAX_THINKING_TOKENS=8000  # 전역 하드 캡 (/effort 위에 우선 적용)',
+        ],
+      },
+      {
+        label: 'Plan mode preemptively (Shift+Tab)',
+        labelKo: 'Plan 모드 선제 사용 (Shift+Tab)',
+        commands: [
+          'Plan mode forces a written plan before any edits — prevents wrong-direction rework',
+          'Rework after a misread spec costs more tokens than 1 extra plan turn',
+        ],
+        commandsKo: [
+          'Plan 모드는 편집 전에 계획을 쓰게 강제 — 잘못된 방향 재작업 방지',
+          '스펙 오독 후 재작업이 plan 1턴 추가보다 훨씬 비쌈',
+        ],
+      },
+      {
+        label: 'Compact or clear when context grows',
+        labelKo: '컨텍스트가 커지면 /compact 또는 /clear',
+        commands: [
+          '/compact  — summarize the session history in place',
+          '/clear    — drop history entirely at a clean task boundary',
+        ],
+        commandsKo: [
+          '/compact — 세션 히스토리를 그 자리에서 요약',
+          '/clear   — 작업 분기점에서 히스토리를 통째로 비우기',
+        ],
+      },
       {
         label: '⚠ Known bug #31640',
         labelKo: '⚠ 알려진 버그 #31640',
@@ -86,14 +128,26 @@ export const ISSUE_MESSAGES = {
       {
         label: 'Avoid opening fresh sessions too often',
         labelKo: '새 세션을 너무 자주 열지 말기',
-        commands: ['Continue the same task in the same session (context switching = cache miss)'],
-        commandsKo: ['같은 작업은 같은 세션에서 계속 (컨텍스트 전환 = 캐시 미스)'],
+        commands: [
+          'Continue the same task in the same session (context switching = cache miss)',
+          'Resume with `claude --continue` instead of starting fresh — preserves the prefix cache',
+        ],
+        commandsKo: [
+          '같은 작업은 같은 세션에서 계속 (컨텍스트 전환 = 캐시 미스)',
+          '새로 시작 대신 `claude --continue`로 재개 — prefix 캐시 보존',
+        ],
       },
       {
         label: 'Stabilize the prompt prefix',
         labelKo: '프롬프트 prefix 안정화',
-        commands: ['System prompts / tool definitions that change per request invalidate the cache every time'],
-        commandsKo: ['요청마다 바뀌는 시스템 프롬프트 / 도구 정의는 매번 캐시를 무효화'],
+        commands: [
+          'System prompts / tool definitions that change per request invalidate the cache every time',
+          'Trim CLAUDE.md — every line ships on every turn; keep only durable rules',
+        ],
+        commandsKo: [
+          '요청마다 바뀌는 시스템 프롬프트 / 도구 정의는 매번 캐시를 무효화',
+          'CLAUDE.md 다이어트 — 모든 줄이 매 턴 실림; 영속적인 규칙만 유지',
+        ],
       },
     ],
   },
@@ -142,6 +196,36 @@ export const ISSUE_MESSAGES = {
           '긴 문서/README 생성 요청은 스크립트로 옮겨 출력 축소',
         ],
       },
+      {
+        label: 'Cap extended-thinking budget (⚠ check /effort)',
+        labelKo: '확장 사고(thinking) 예산 제한 (⚠ /effort 확인)',
+        commands: [
+          '⚠ Check `/effort` — `xhigh` burns thinking tokens that count as output',
+          '/effort medium  — Anthropic-official default; switch back from xhigh',
+          'export MAX_THINKING_TOKENS=8000  # global hard cap',
+        ],
+        commandsKo: [
+          '⚠ `/effort` 확인 — `xhigh`는 thinking 토큰을 다량 소비 (출력으로 집계)',
+          '/effort medium  — Anthropic 공식 기본값; xhigh에서 복귀',
+          'export MAX_THINKING_TOKENS=8000  # 전역 하드 캡',
+        ],
+      },
+      {
+        label: 'Model matching strategy (80/15/5 rule)',
+        labelKo: '모델 매칭 전략 (80/15/5 비율)',
+        commands: [
+          'Sonnet 80% — daily coding, edits, refactors',
+          'Opus 15%  — complex design, multi-file refactor planning, deep debugging',
+          'Haiku 5%  — boilerplate, labeling, summaries, subagent work',
+          'Switch with `/model sonnet` / `/model opus` / `/model haiku`',
+        ],
+        commandsKo: [
+          'Sonnet 80% — 일상 코딩, 편집, 리팩터링',
+          'Opus 15%  — 복잡한 설계, 다파일 리팩터 계획, 깊은 디버깅',
+          'Haiku 5%  — 보일러플레이트, 라벨링, 요약, 서브에이전트 작업',
+          '`/model sonnet` / `/model opus` / `/model haiku`로 전환',
+        ],
+      },
     ],
   },
   HIGH_REQUEST_COUNT: {
@@ -165,6 +249,40 @@ export const ISSUE_MESSAGES = {
         commands: ['Check that the agent is not repeating the same test/search in a loop'],
         commandsKo: ['에이전트가 동일한 테스트/검색을 루프로 반복하고 있지 않은지 확인'],
       },
+      {
+        label: 'Delegate large searches to a subagent (Sonnet by default)',
+        labelKo: '대형 검색은 서브에이전트로 위임 (Sonnet 기본)',
+        commands: [
+          'Use the Task tool with the Explore subagent for repo-wide searches',
+          'Pin custom subagents to Sonnet/Haiku in their frontmatter: `model: sonnet`',
+          'The subagent runs in its own context — main session keeps a clean prefix',
+        ],
+        commandsKo: [
+          '레포 전반 검색은 Task 도구의 Explore 서브에이전트로 위임',
+          '커스텀 서브에이전트는 frontmatter에 `model: sonnet` 명시 (Opus 자동 상속 방지)',
+          '서브에이전트는 자체 컨텍스트에서 실행 — 메인 세션 prefix가 깨끗하게 유지됨',
+        ],
+      },
+      {
+        label: 'Migrate MCP servers → Skills / CLI',
+        labelKo: 'MCP 서버 → Skills / CLI 전환',
+        commands: [
+          '30 MCP tools ≈ 3,600 tokens loaded every turn — measured by mcp2cli (HN 146pts)',
+          'Replacing rarely-used MCPs with Skills (loaded on demand) or shell CLI cuts ~96%',
+          'Audit `.mcp.json` and disable everything not used weekly',
+        ],
+        commandsKo: [
+          'MCP 도구 30개 ≈ 매 턴 3,600 토큰 상시 로드 (mcp2cli 측정, HN 146pts)',
+          '드물게 쓰는 MCP를 Skills(필요 시 로드) 또는 셸 CLI로 대체하면 ~96% 절감',
+          '`.mcp.json` 점검해 주간에 안 쓰는 건 모두 비활성화',
+        ],
+      },
+      {
+        label: 'Trim hooks that inject context',
+        labelKo: '컨텍스트 주입 훅 정리',
+        commands: ['Disable noisy PreToolUse/PostToolUse hooks unless they earn their tokens'],
+        commandsKo: ['값을 못 하는 PreToolUse/PostToolUse 훅은 비활성화'],
+      },
     ],
   },
   FREQUENT_CACHE_REBUILD: {
@@ -182,6 +300,30 @@ export const ISSUE_MESSAGES = {
         commands: ['Continue one task in one Claude Code session'],
         commandsKo: ['하나의 작업은 하나의 Claude Code 세션에서 계속'],
       },
+      {
+        label: 'Use /compact at task boundaries',
+        labelKo: '작업 분기점에서 /compact 사용',
+        commands: [
+          '/compact summarizes prior turns in place — keeps the session alive without re-reading everything',
+          'Better than starting a fresh session, which forces a full prefix rebuild',
+        ],
+        commandsKo: [
+          '/compact는 이전 턴을 그 자리에서 요약 — 전체 재읽기 없이 세션 유지',
+          '새 세션을 여는 것보다 나음 (새 세션은 prefix 전체 재구축 강제)',
+        ],
+      },
+      {
+        label: 'Avoid re-reading large files',
+        labelKo: '큰 파일 반복 Read 피하기',
+        commands: [
+          'Prefer `git diff` to see what changed instead of re-reading the whole file',
+          'Edit returns success silently — no need to Read after editing',
+        ],
+        commandsKo: [
+          '파일 전체 다시 읽기 대신 `git diff`로 변경분만 확인',
+          'Edit는 성공 시 조용히 끝남 — 편집 후 Read 다시 할 필요 없음',
+        ],
+      },
     ],
   },
 };
@@ -197,28 +339,28 @@ export const ISSUE_MESSAGES = {
  */
 export const ISSUE_TIPS = {
   LARGE_INPUT_PER_REQUEST: {
-    en: 'Disable 1M context: `export CLAUDE_CODE_DISABLE_1M_CONTEXT=1` (or ⌥P toggle)',
-    ko: '1M 컨텍스트 끄기: `export CLAUDE_CODE_DISABLE_1M_CONTEXT=1` (또는 ⌥P 토글)',
+    en: 'Check `/effort` — `xhigh` is the #1 cap killer; switch to `/effort medium` (or `low`); disable 1M context; `/compact` when context grows',
+    ko: '`/effort` 확인 — `xhigh`가 캡 소진 1순위 원인, `medium`(또는 `low`)으로 복귀; 1M 컨텍스트 끄기; 컨텍스트 커지면 `/compact`',
   },
   LOW_HIT_RATE: {
-    en: 'Continue same task in same session; keep prompt prefix stable',
-    ko: '같은 작업은 같은 세션에서 계속; 프롬프트 prefix 안정 유지',
+    en: 'Continue with `claude --continue`; keep CLAUDE.md trim — every line ships every turn',
+    ko: '`claude --continue`로 재개; CLAUDE.md 다이어트 — 모든 줄이 매 턴 실림',
   },
   BUCKET_5M_DOMINANT: {
     en: 'Send any prompt within 5min to keep cache warm; Max plan unlocks 1h TTL',
     ko: '5분 이내 한 번 더 보내 캐시 유지; Max 플랜은 1시간 TTL 제공',
   },
   HIGH_OUTPUT_RATIO: {
-    en: 'Avoid full-file rewrites — prefer Edit tool; move long generations to scripts',
-    ko: '파일 전체 재작성 피하기 — Edit 도구 사용; 긴 생성은 스크립트로',
+    en: 'Check `/effort` (`xhigh` inflates output); prefer Edit over full rewrites; model matching: Sonnet 80% / Opus 15% / Haiku 5%',
+    ko: '`/effort` 확인 (`xhigh`는 출력 폭증); Edit 도구 우선 (전체 재작성 피하기); 모델 매칭: Sonnet 80% / Opus 15% / Haiku 5%',
   },
   HIGH_REQUEST_COUNT: {
-    en: 'Bundle independent calls into one message; check for retry loops',
-    ko: '독립 호출은 한 메시지에 묶기; 재시도 루프 점검',
+    en: 'Bundle calls into one message; delegate to subagents (`model: sonnet`); migrate MCP→Skills/CLI (30 tools = ~3,600 tokens/turn)',
+    ko: '호출은 한 메시지에 묶기; 서브에이전트(`model: sonnet`)로 위임; MCP→Skills/CLI 전환 (30 tools ≈ 매 턴 3,600 토큰)',
   },
   FREQUENT_CACHE_REBUILD: {
-    en: 'Continue one task in one session — short sessions trigger rebuild',
-    ko: '한 작업은 한 세션에서 — 짧은 세션은 재빌드 유발',
+    en: 'Continue one task in one session; use `/compact` at boundaries; avoid re-reading large files (use `git diff`)',
+    ko: '한 작업은 한 세션에서; 분기점에선 `/compact`; 큰 파일 재읽기 대신 `git diff`',
   },
 };
 

package/src/caps-cache.js

@@ -1,7 +1,7 @@
 /**
  * Statusline snapshot cache — the rate-limit numbers and model name only flow
  * through stdin from Claude Code's statusline contract, but we want the table
- * view (e.g. invoked by `/token-monitor`) to surface the same data. So
+ * view (`claude-token-saver --days N`) to surface the same data. So
  * whenever the statusline path sees them it writes them here, and the table
  * path reads them back if its own stdin was empty.
  *

package/src/installer.js

@@ -1,7 +1,11 @@
 /**
- * Installs the Claude Code integration assets:
+ * Installs the Claude Code integration asset:
  *   - Skill:  ~/.claude/skills/claude-token-saver/SKILL.md
- *   - Slash:  ~/.claude/commands/token-monitor.md
+ *
+ * v2.6.0 consolidates `/token-monitor` into the skill (was redundant with the
+ * auto-trigger). On install we actively remove a legacy
+ * ~/.claude/commands/token-monitor.md if present so users don't see two
+ * overlapping entry points.
  *
  * All paths are resolved with node:path so Windows backslashes and POSIX
  * forward-slashes are both handled. Directories are created with
@@ -9,7 +13,7 @@
  * exist on every platform.
  */
 
-import { writeFileSync, mkdirSync, existsSync } from 'node:fs';
+import { writeFileSync, mkdirSync, existsSync, unlinkSync } from 'node:fs';
 import { join } from 'node:path';
 import { claudeUserDir } from './paths.js';
 
@@ -36,6 +40,9 @@ countdown, savings, and (when relevant) a leading warning chip.
   \`claude-token-saver handoff\`).
 - The user wants to see the token-usage history file or asks for a summary
   of recent warnings.
+- The user asks for a quick token report or "current state" check (the
+  skill replaces the legacy \`/token-monitor\` slash command — same workflow,
+  triggered by intent rather than a typed slash).
 
 ## What to do
 
@@ -97,41 +104,6 @@ History files live under the OS-appropriate user-data dir:
 Each day's file is plain Markdown — safe to open in any editor.
 `;
 
-const COMMAND_BODY = `---
-description: Show recent claude-token-saver warning history and a fresh report.
----
-
-You are responding to the \`/token-monitor\` slash command. The user wants a
-quick read of their Claude Code token usage and any active warnings — most
-importantly: **what just happened, and how do I handle it?**
-
-Steps:
-
-1. **Lead with the most recent warning.** Run \`claude-token-saver last\`
-   first and surface its output verbatim (or lightly summarized) at the top
-   of your reply. This returns the latest warning event (chip + detail +
-   timestamp) followed by the full advice block. If \`last\` says no recent
-   warnings, mention that and skip ahead — you can stop here unless the user
-   asked for more.
-2. Run \`claude-token-saver history --days 7\` and capture the output. Use it
-   only to add context — e.g. "this is the 3rd cache miss today" — not to
-   re-print the whole file. Each entry is bilingual and includes a \`💡\`
-   action tip inline.
-3. Run \`claude-token-saver --days 1\` and capture the output for any extra
-   color you want to add: TTL breakdown, cost impact, daily trend, or active
-   spikes. Skip if step 1 already covered what the user needs.
-4. Summarize for the user:
-   - **What just fired** — the chip + the time + a sentence on what caused it
-     (from \`last\`).
-   - **What to do** — the action tip from \`last\`. For cap-warn (\`🚨 5H/7D NN%\`),
-     surface \`claude-token-saver handoff\` prominently so they can back up
-     state before the cap blocks them.
-   - **Today's pattern** (optional) — when warnings cluster in time, mention it.
-
-Keep the summary to ~10 lines. The user can re-run the underlying commands
-themselves for the full output.
-`;
-
 function writeIfNeeded(file, body, force) {
   const existed = existsSync(file);
   if (existed && !force) return { path: file, action: 'exists' };
@@ -146,16 +118,18 @@ export function installSkill({ force = false } = {}) {
   return writeIfNeeded(file, SKILL_BODY, force);
 }
 
-export function installCommand({ force = false } = {}) {
-  const dir = join(claudeUserDir(), 'commands');
-  const file = join(dir, 'token-monitor.md');
-  mkdirSync(dir, { recursive: true });
-  return writeIfNeeded(file, COMMAND_BODY, force);
+// Removes the legacy /token-monitor slash command from prior versions.
+// v2.6.0 consolidated it into the skill — the file would otherwise linger.
+export function removeLegacyCommand() {
+  const file = join(claudeUserDir(), 'commands', 'token-monitor.md');
+  if (!existsSync(file)) return { path: file, action: 'absent' };
+  unlinkSync(file);
+  return { path: file, action: 'removed' };
 }
 
 export function installAll({ force = false } = {}) {
   return {
     skill: installSkill({ force }),
-    command: installCommand({ force }),
+    legacy: removeLegacyCommand(),
   };
 }