claude-usage-limits 1.39.2 → 1.39.4

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "usage-limits",
   "displayName": "Usage Limits",
-  "version": "1.39.2",
+  "version": "1.39.4",
   "description": "Puts your remaining Claude Code usage limit into Claude's context before every prompt, so it opens with what fits in the budget instead of starting work that gets cut off. Reports headroom as turns rather than percentages, prices a job before you start it, and detects your plan tier.",
   "author": {
     "name": "Ridelink",

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "usage-limits",
-  "version": "1.39.2",
+  "version": "1.39.4",
   "description": "Reports how much of your Codex usage limit is left as turns of work rather than a percentage, prices a job before you start it, and counts the other agents sharing the same budget.",
   "author": {
     "name": "Ridelink",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-usage-limits",
-  "version": "1.39.2",
+  "version": "1.39.4",
   "description": "Puts your remaining Claude Code usage limit into Claude's context before every prompt, so it opens with what fits in the budget instead of starting work that gets cut off. Reports headroom as turns rather than percentages, prices a job before you start it, and detects your plan tier.",
   "keywords": [
     "claude",

package/skills/usage-limits/scripts/agy-hook.js

@@ -33,7 +33,12 @@ const usage = require('./usage.js');
 const host = require('./host.js');
 const mode = require('./mode.js');
 const ceiling = require('./ceiling.js');
-const brief = require('./brief.js');
+// brief.js is loaded only on PreInvocation: PreToolUse runs before tool calls
+// and must stay a few small requires, not the whole brief.
+
+// PreToolUse must always name a decision. Antigravity reads an answer with no
+// decision as a refusal, so the old bare {} blocked every tool it ran on.
+const ALLOW = { decision: 'allow' };
 
 const EVENTS = ['PreInvocation', 'PostInvocation', 'PreToolUse', 'PostToolUse', 'Stop'];
 
@@ -116,14 +121,14 @@ async function run(now, input, argv) {
   usage.setHost(host.GEMINI);
   const sessionId = input && input.conversationId ? String(input.conversationId) : null;
   const budget = mode.forSession({ sessionId });
-  if (budget.policy.briefStyle === 'none') return {};
+  if (budget.policy.briefStyle === 'none') return event === 'PreToolUse' ? ALLOW : {};
 
   if (event === 'PreToolUse') {
     const tool = toolNameOf(input);
-    if (!ceiling.isMultiplier(tool)) return {};
+    if (!ceiling.isMultiplier(tool)) return ALLOW;
     const at = ceiling.assess(Object.assign({ state: budget.state, env: process.env, sessionId }, reading(percentNow(now))));
     const call = ceiling.verdict(at, tool);
-    if (call.decision !== 'deny') return {};
+    if (call.decision !== 'deny') return ALLOW;
     return { decision: 'deny', reason: call.reason };
   }
 
@@ -133,7 +138,7 @@ async function run(now, input, argv) {
     // transient system message, which is exactly what a per-turn figure is.
     let text = '';
     try {
-      text = await brief.run(now, input);
+      text = await require('./brief.js').run(now, input, { host: host.GEMINI });
     } catch (err) {
       text = '';
     }
@@ -162,11 +167,13 @@ if (require.main === module) {
       },
       () => {
         // Hooks block the agent loop here, so a failure has to be silent and
-        // well formed rather than loud.
-        process.stdout.write('{}\n');
+        // well formed rather than loud - and on PreToolUse, well formed means
+        // an explicit allow, never a bare {} that reads as a refusal.
+        const pre = process.argv.slice(2).join(' ').includes('--event PreToolUse');
+        process.stdout.write(JSON.stringify(pre ? ALLOW : {}) + '\n');
         process.exit(0);
       }
     );
 }
 
-module.exports = { EVENTS, eventFrom, toolNameOf, percentNow, run };
+module.exports = { EVENTS, ALLOW, eventFrom, toolNameOf, percentNow, run };

package/skills/usage-limits/scripts/brief.js

@@ -686,11 +686,19 @@ function briefText(input) {
   if (Number.isFinite(parts.turnsLeft)) {
     // Another session spending the same budget means fewer of those turns are
     // yours, so say both numbers rather than the flattering one.
+    // The stance below is judged on yourTurnsLeft, so whenever that is the
+    // smaller number it has to be on the line too. Printing only the window's
+    // figure is how a brief came to say "the budget is nearly gone" beside
+    // "about 594 turns of headroom": both were true, of different numbers.
+    const yours = Number.isFinite(parts.yourTurnsLeft) ? parts.yourTurnsLeft : null;
+    const fewerYours = yours !== null && yours < parts.turnsLeft * 0.75;
     const shared =
-      parts.sessions > 1 && Number.isFinite(parts.yourTurnsLeft)
-        ? ' (' + parts.sessions + ' sessions active, roughly ' + parts.yourTurnsLeft +
+      parts.sessions > 1 && yours !== null
+        ? ' (' + parts.sessions + ' sessions active, roughly ' + yours +
           ' of them yours)'
-        : '';
+        : fewerYours
+          ? ' (about ' + count(yours, 'turn') + ' of that yours at this context size)'
+          : '';
     bound.push('about ' + count(parts.turnsLeft, 'turn') + ' of headroom' + shared);
     // A turn count is a poor sense of urgency when several agents are spending
     // at once: two hundred turns sounds like plenty and can be gone in ten
@@ -1411,7 +1419,12 @@ function fastModeFor(sessionId) {
   }
 }
 
-async function run(now, hookInput) {
+const GEMINI_UNREADABLE =
+  "[usage-limits] Antigravity's quota is not readable by this plugin, so there is no budget figure and no fan-out ceiling here; Antigravity's own usage panel has the number.";
+const LIMIT_REACHED =
+  '[usage-limits] The meter says the usage limit is reached. Start nothing new: save the work, write the hand-off note in this turn, and end the turn.';
+
+async function run(now, hookInput, opts) {
   if (String(process.env.USAGE_LIMITS_BRIEF || '').toLowerCase() === 'off') return '';
 
   const sessionId = hookInput && (hookInput.session_id || hookInput.conversationId) ? (hookInput.session_id || hookInput.conversationId) : null;
@@ -1424,15 +1437,21 @@ async function run(now, hookInput) {
   const budget = mode.forSession({ sessionId });
   if (budget.policy.briefStyle === 'none') return guardLine(now, budget);
 
-  // Settle host here, before any file is read.
-  const detectedHost = host.detect(process.argv.slice(2), process.env);
-  const effectiveHost = (hookInput && (hookInput.conversationId || hookInput.invocationNum !== undefined))
-    ? host.GEMINI
-    : detectedHost;
-  usage.setHost(effectiveHost);
+  // Settle host here, before any file is read. A caller that already knows the
+  // host (agy-hook runs only inside Antigravity) says so, and that wins over
+  // detection: an empty or unparseable payload used to fall through to
+  // Claude Code and put Claude's usage, model and session into Antigravity.
+  const detectedHost = host.detectFromHook(process.argv.slice(2), process.env, hookInput);
+  const effectiveHost = (opts && opts.host)
+    || ((hookInput && (hookInput.conversationId || hookInput.invocationNum !== undefined))
+      ? host.GEMINI
+      : detectedHost);
+  const gemini = usage.setHost(effectiveHost) === host.GEMINI;
   // A prompt has arrived, so this session is working, and the prompt itself
-  // says whether it asked for ultracode. The panel animates from this.
-  activity.mark(
+  // says whether it asked for ultracode. The panel animates from this. The
+  // activity file is Claude Code's, so an Antigravity conversation is never
+  // counted in it.
+  if (!gemini) activity.mark(
     'working',
     sessionId,
     Object.assign(
@@ -1469,7 +1488,11 @@ async function run(now, hookInput) {
   // one quick failure and then a widening backoff, never a wait on every
   // prompt; USAGE_LIMITS_FETCH=off turns it off.
   try {
-    if (usage.isCodex()) {
+    if (gemini) {
+      // Antigravity publishes no meter this plugin can read, and refreshing
+      // Claude's /usage reading from inside Antigravity spends a request on
+      // the wrong account.
+    } else if (usage.isCodex()) {
       // Codex only writes its meter when it makes a request, so between turns
       // the newest figure can be half an hour old. Ask it, the way /status
       // does, when the reading has aged.
@@ -1492,7 +1515,13 @@ async function run(now, hookInput) {
     // The reading on disk is still there.
   }
   const base = usage.collect(now);
-  if (!base.utilization) return '';
+  if (!base.utilization) {
+    if (gemini) return sayOnce(sessionId, GEMINI_UNREADABLE, now);
+    // Codex can say a limit is already hit while giving no percentage; that
+    // is the plainest reading there is, not a missing one.
+    if (base.reachedType || base.spendControlReached) return sayOnce(sessionId, LIMIT_REACHED, now);
+    return '';
+  }
 
   const all = readCache();
   // Which setting the fit question was already asked for, read BEFORE anything

package/skills/usage-limits/scripts/ceiling.js

@@ -50,7 +50,8 @@ const MULTIPLIERS = [
 // Antigravity derives tool names by lowercasing the step type and stripping
 // CORTEX_STEP_TYPE_, so its names are snake_case and cannot be matched by the
 // list above.
-const MULTIPLIER_PATTERN = /^(agent|task|workflow|subagent|dispatch|spawn_[a-z_]*agent|run_[a-z_]*agent)$/i;
+// invoke_subagent and browser_subagent are the two Antigravity fan-out tools.
+const MULTIPLIER_PATTERN = /^(agent|task|workflow|subagent|dispatch|spawn_[a-z_]*agent|run_[a-z_]*agent|invoke_subagent|browser_subagent)$/i;
 
 // How far below the ceiling the warning starts. Ten points is about one long
 // turn at xhigh on a 5-hour window, which is the last moment a warning can

package/skills/usage-limits/scripts/codex.js

@@ -777,6 +777,30 @@ function liveFile() {
   return path.join(homeDir(), 'usage-limits-codex-live.json');
 }
 
+// The app-server's account/rateLimits/read answers in camelCase (usedPercent,
+// windowDurationMins, resetsAt, planType ...) while rollouts and every reader
+// here use snake_case. Stored as received, a fresh live reading was invisible
+// to readingsOf() and the brief, pulse and ceiling all went blind on Codex.
+// One conversion, applied where a reading is taken and again where the file
+// is read back, because a camelCase file may already be on disk.
+const METER_RENAMES = { window_duration_mins: 'window_minutes' };
+function snakeKey(key) {
+  const snake = String(key).replace(/[A-Z]/g, (c) => '_' + c.toLowerCase());
+  return METER_RENAMES[snake] || snake;
+}
+function normalizeMeter(meter) {
+  if (Array.isArray(meter)) return meter.map(normalizeMeter);
+  if (!meter || typeof meter !== 'object') return meter;
+  const out = {};
+  for (const [key, value] of Object.entries(meter)) {
+    const name = snakeKey(key);
+    // A snake_case field that is already there wins over a converted twin.
+    if (Object.prototype.hasOwnProperty.call(out, name) && name !== key) continue;
+    out[name] = normalizeMeter(value);
+  }
+  return out;
+}
+
 function readLiveMeter() {
   let parsed;
   try {
@@ -786,7 +810,7 @@ function readLiveMeter() {
   }
   if (!parsed || typeof parsed !== 'object') return null;
   if (!Number.isFinite(parsed.at) || !parsed.meter || typeof parsed.meter !== 'object') return null;
-  return parsed;
+  return Object.assign({}, parsed, { meter: normalizeMeter(parsed.meter) });
 }
 
 function writeLiveMeter(reading) {
@@ -1045,7 +1069,7 @@ function refresh(options) {
             return;
           }
           const limits = message.result && message.result.rateLimits;
-          finish(null, { at: Date.now(), meter: limits || null });
+          finish(null, { at: Date.now(), meter: limits ? normalizeMeter(limits) : null });
           return;
         }
       }
@@ -1073,6 +1097,7 @@ module.exports = {
   SLOTS,
   WEIGHTS,
   PLANS,
+  normalizeMeter,
   homeDir,
   sessionsDir,
   findExecutable,

package/skills/usage-limits/scripts/host.js

@@ -112,7 +112,21 @@ function detect(argv, env) {
   return CLAUDE;
 }
 
+// Codex hook payloads always carry turn_id; Claude Code's never do. Codex also
+// loads an installed plugin's Claude-style hooks/hooks.json, which runs these
+// scripts with no --host and may set CLAUDE_PLUGIN_ROOT, so detect() alone
+// called that Codex turn Claude Code and briefed Claude's budget into Codex.
+// An explicit --host or USAGE_LIMITS_HOST still wins.
+function detectFromHook(argv, env, input) {
+  const args = argv || [];
+  const environment = env || process.env;
+  if (args.includes('--host') || normalise(environment.USAGE_LIMITS_HOST)) return detect(args, environment);
+  if (input && typeof input === 'object' && input.turn_id !== undefined && input.turn_id !== null) return CODEX;
+  return detect(args, environment);
+}
+
 module.exports = {
+  detectFromHook,
   CLAUDE,
   CODEX,
   GEMINI,

package/skills/usage-limits/scripts/install-antigravity.js

@@ -77,7 +77,9 @@ function hooks() {
     'usage-limits-ceiling': {
       PreToolUse: [
         {
-          matcher: '.*',
+          // Only the fan-out tools: a node start before every tool call cost
+          // 3-6 s each here, and ordinary tools have nothing to ask the ceiling.
+          matcher: 'invoke_subagent|browser_subagent',
           hooks: [{ type: 'command', command: hookCommand('PreToolUse'), timeout: 10 }],
         },
       ],

package/skills/usage-limits/scripts/pulse.js

@@ -225,7 +225,7 @@ async function run(now, hookInput) {
   const budget = mode.forSession({ sessionId });
   if (budget.policy.refreshSeconds === 0) return '';
 
-  usage.setHost(host.detect(process.argv.slice(2), process.env));
+  usage.setHost(host.detectFromHook(process.argv.slice(2), process.env, hookInput));
 
   // A subagent finishing is the other reason to look, and on a busy afternoon
   // it is the more important one.