@rex_koh/subagent-budget-guard 0.2.0 → 0.3.0

package/.claude-plugin/plugin.json

@@ -2,7 +2,7 @@
   "name": "agent-guard",
   "displayName": "Agent Guard",
   "description": "Hard-deny subagent launches, record verified subagent usage, and enforce a session budget against Claude Code's 5-hour rate-limit percentage.",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "author": {
     "name": "ClaudeSubAgentSuppressor"
   },

package/README.md

@@ -10,16 +10,9 @@ Recommended Claude Code install:
 /plugin marketplace add rexkoh425/ClaudeSubAgentSuppressor
 /plugin install agent-guard@subagent-budget-tools
 /agent-guard:init
-/agent-guard:doctor
 ```
 
-After `/agent-guard:init`, fully exit and reopen Claude Code before verification so the statusLine bridge from `settings.json` is active. Some Claude Code builds do not provide an in-session plugin reload command.
-
-Useful after install:
-
-```text
-/agent-guard:status
-```
+After `/agent-guard:init`, fully exit and reopen Claude Code so the statusLine bridge from `settings.json` is active. Some Claude Code builds do not provide an in-session plugin reload command.
 
 ## NPM Package
 
@@ -30,6 +23,7 @@ Claude Code plugin discovery is marketplace-based, so npm is mainly useful as a
 ```bash
 npm install -g @rex_koh/subagent-budget-guard
 agent-guard doctor --offline
+agent-guard status
 ```
 
 Maintainer publish command:
@@ -57,7 +51,7 @@ enforcement_enabled=true
 
 For existing installs, setup also removes obsolete `max_subagents_per_session` and `max_agent_team_tasks_per_session` options from this plugin's Claude settings.
 
-The setup skill can also ask for custom values. For direct terminal setup, use:
+The setup skill can ask for custom values. For direct terminal setup, use:
 
 ```bash
 agent-guard init

package/lib/guard.js

@@ -676,10 +676,7 @@ export async function handleUserPromptSubmit(input, env = process.env) {
   const sessionId = input?.session_id || 'unknown-session';
   const config = loadConfig(env);
   const state = await readState(sessionId, env);
-  const tokenBudgetReason = subagentTokenBudgetDecision(state, config, {
-    includeWarning: false
-  });
-  const reason = tokenBudgetReason?.reason || fiveHourBudgetDecision(state, config);
+  const reason = fiveHourBudgetDecision(state, config);
 
   if (!reason) {
     return { exitCode: 0, stdout: null, stderr: '' };

package/lib/verifier.js

@@ -111,7 +111,7 @@ export async function runOfflineVerification({
         entry.source?.package === '@rex_koh/subagent-budget-guard',
         'marketplace npm package mismatch'
       );
-      assert(entry.source?.version === '0.2.0', 'marketplace npm version mismatch');
+      assert(entry.source?.version === '0.3.0', 'marketplace npm version mismatch');
       return marketplacePath;
     });
   } else {
@@ -175,12 +175,7 @@ export async function runOfflineVerification({
       'bin/verify.js',
       'lib/guard.js',
       'lib/verifier.js',
-      'skills/init/SKILL.md',
-      'skills/status/SKILL.md',
-      'skills/doctor/SKILL.md',
-      'skills/setup/SKILL.md',
-      'skills/report/SKILL.md',
-      'skills/verify/SKILL.md'
+      'skills/init/SKILL.md'
     ];
     for (const script of scripts) {
       assert(await pathExists(path.join(root, script)), `missing ${script}`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rex_koh/subagent-budget-guard",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Claude Code plugin that blocks subagents by default, records verified subagent usage, and enforces 5-hour usage budgets.",
   "license": "MIT",
   "author": "ClaudeSubAgentSuppressor",
@@ -32,11 +32,7 @@
     "LICENSE"
   ],
   "bin": {
-    "agent-guard": "bin/agent-guard.js",
-    "ag": "bin/agent-guard.js",
-    "subagent-budget-guard-report": "bin/report.js",
-    "subagent-budget-guard-setup": "bin/setup.js",
-    "subagent-budget-guard-verify": "bin/verify.js"
+    "agent-guard": "bin/agent-guard.js"
   },
   "publishConfig": {
     "access": "public"

package/skills/init/SKILL.md

@@ -35,7 +35,9 @@ node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" init \
   --config enforcement_enabled=<true-or-false>
 ```
 
-Then tell the user to fully exit and reopen Claude Code, interact once so the statusLine bridge receives fresh session JSON, and run:
+Then tell the user to fully exit and reopen Claude Code, then interact once so the statusLine bridge receives fresh session JSON.
+
+For optional terminal verification, run:
 
 ```bash
 node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --live

package/skills/doctor/SKILL.md

@@ -1,20 +0,0 @@
----
-description: Verify Agent Guard installation and offline behavior.
-disable-model-invocation: true
----
-
-# Doctor Agent Guard
-
-For default offline verification, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --offline
-```
-
-For live local installation checks, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --live
-```
-
-The live verifier does not submit Claude prompts. It checks local plugin shape, `claude plugin validate` when available, plugin listing shape, and statusLine bridge setup.

package/skills/report/SKILL.md

@@ -1,20 +0,0 @@
----
-description: Compatibility alias for /agent-guard:status.
-disable-model-invocation: true
----
-
-# Report Agent Guard Usage
-
-Prefer `/agent-guard:status`.
-
-Run this command:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" status
-```
-
-If the user asks for machine-readable output, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" status --json
-```

package/skills/setup/SKILL.md

@@ -1,44 +0,0 @@
----
-description: Compatibility alias for /agent-guard:init.
----
-
-# Setup Agent Guard
-
-Prefer `/agent-guard:init`.
-
-Ask the user whether to use the recommended defaults or customize the values. If they choose defaults, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" init --defaults
-```
-
-If they choose custom values, ask for each value below. The value in parentheses is the default; accept a blank answer as the default.
-
-```text
-max_concurrent_subagents=1
-max_subagent_tokens_per_session=100000
-subagent_token_warning_threshold_percent=95
-session_five_hour_budget_percent=25
-absolute_five_hour_ceiling_percent=95
-enforcement_enabled=true
-```
-
-Then run setup with the chosen values:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" init \
-  --config max_concurrent_subagents=<value> \
-  --config max_subagent_tokens_per_session=<value> \
-  --config subagent_token_warning_threshold_percent=<value> \
-  --config session_five_hour_budget_percent=<value> \
-  --config absolute_five_hour_ceiling_percent=<value> \
-  --config enforcement_enabled=<true-or-false>
-```
-
-Then tell the user to fully exit and reopen Claude Code, interact once so the statusLine bridge receives fresh session JSON, and run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --live
-```
-
-The live verifier does not submit Claude prompts. It checks local plugin shape, Claude plugin validation when `claude` is on `PATH`, and whether the statusLine bridge is configured.

package/skills/status/SKILL.md

@@ -1,18 +0,0 @@
----
-description: Show the current Agent Guard report.
-disable-model-invocation: true
----
-
-# Status Agent Guard
-
-Run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" status
-```
-
-If the user asks for machine-readable output, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" status --json
-```

package/skills/verify/SKILL.md

@@ -1,22 +0,0 @@
----
-description: Compatibility alias for /agent-guard:doctor.
-disable-model-invocation: true
----
-
-# Verify Agent Guard
-
-Prefer `/agent-guard:doctor`.
-
-For default offline verification, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --offline
-```
-
-For live local installation checks, run:
-
-```bash
-node "${CLAUDE_PLUGIN_ROOT}/bin/agent-guard.js" doctor --live
-```
-
-The live verifier does not submit Claude prompts. It checks local plugin shape, `claude plugin validate` when available, plugin listing shape, and statusLine bridge setup.