Skip to content
ccmods.app

agent-usage-guard

v1.0.3

Holds subagent fan-out, heavy-context prompts and retry loops before they spend your 5-hour or weekly plan usage.

by @rbartoli2Updated Broad accessGitHub

Installation

  1. Install from your shell (answer y to add the marketplace)

    $ claude plugin install agent-usage-guard --marketplace rbartoli/agent-usage-guard

Already in a session? Type this at the Claude Code prompt:

$ /plugin install agent-usage-guard --marketplace rbartoli/agent-usage-guard

Pick a scope and the mod is active right away. Needs Claude Code 2.1.287 or later, in a terminal. Full install guide

What does agent-usage-guard do?

  • Asks before new agents once a plan window is 80% full and refuses them at 95%
  • Caps agents at 4 running at once and 12 starts per 10 minutes across every session on the machine
  • Asks before a prompt that would re-read 500k tokens: compact then send, send anyway, or cancel
  • Refuses the identical retry after 3 identical tool failures
  • /agent-guard shows status, with allow, pause, resume and report subcommands

What do you need?

  • Claude Code 2.1.287 or later (terminal)

What can this mod access?

Read from the mod's code with claude plugin validate on Claude Code 2.1.296, synced Oct 12, 2026. This is a capability list, not a security audit. Mods run with your permissions, so install only from authors you trust.

Broad access
  • Submits prompts or messages for you
  • Intercepts or rewrites tool calls
  • Reads or rewrites your prompts
  • Reads environment variables or settings
  • Reads your conversation
Hooks and engine calls (16 + 17)

hooks

  • agent.spawn
  • classic.PostCompact
  • classic.PostToolUse
  • classic.SessionStart
  • classic.Stop
  • classic.StopFailure
  • classic.SubagentStart
  • classic.UserPromptSubmit
  • command.run{command=agent-guard}
  • prompt.submit
  • session.end
  • session.measure
  • session.start
  • tool.call
  • turn.complete
  • turn.step

calls

  • $.clock.after
  • $.clock.now
  • $.command.register
  • $.env.get
  • $.prompt.fill
  • $.prompt.read
  • $.prompt.submit
  • $.session.compact
  • $.session.id
  • $.session.usage
  • $.store.delete
  • $.store.get
  • $.store.keys
  • $.store.set
  • $.ui.ask
  • $.ui.log
  • $.ui.status

How do you uninstall agent-usage-guard?

$ claude plugin uninstall agent-usage-guard

To switch it off without removing it, use claude plugin disable with the same name.

Do you maintain agent-usage-guard?

Add this badge to your README so users can jump to this page for the install command and access report.

Listed on ccmods.app
[![Listed on ccmods.app](https://ccmods.app/badge.svg)](https://ccmods.app/mods/agent-usage-guard)

Related mods

jev-pilot

Akramovic1/jev-pilot

10

Asks the Jev decision model before each turn to set reasoning effort, subagent model and the one skill a prompt needs.

UIPromptToolsBroad access
$ claude plugin install jev-pilot --marketplace Akramovic1/jev-pilot

crew-chief

aykyusuf/crew-chief

2

Routes subagent work by difficulty across seven Haiku, Sonnet and Opus tiers and offers to cap subagent effort as plan limits fill.

PromptCommandsModerate access
$ claude plugin install crew-chief --marketplace aykyusuf/crew-chief

cadence

crenshawdev/cadence

8

Plan, execute and verify workflow kept in .planning/, with a band and pane showing the current phase, running agents, token spend and cache breaks.

UIPromptToolsBroad access
$ claude plugin install cadence --marketplace crenshawdev/cadence