Claude Code, Codex and ChatGPT Work · Plugin 0.3.3

Your agents, stopped at the limit

Check agent actions against your policy, stop runaway spawns and keep the signed record on your machine.

Install free What leaves your machine

Opening card of the recorded fan-out replay: Claude Code launched 15 copies in 36.6 seconds, then the copies tried to launch more, 8 attempts refused. — A deliberate fan-out test on one of our machines, Sep 23 2026, replayed at 16x from the signed ledger and the fifteen child transcripts: fifteen copies in 36.6 seconds, the next three attempts refused at the ceiling, then five more refused when the copies tried to launch their own, 29,997,622 tokens at $78.21 list price (not a bill). The opening card states the result; every time on screen is recorded time; the token fill is drawn between each copy's recorded start and end. The animated card on /try is a prepared session; this clip is not.

Opening card of the recorded fan-out replay, phone layout: Claude Code launched 15 copies in 36.6 seconds, then the copies tried to launch more, 8 attempts refused. — A deliberate fan-out test on one of our machines, Sep 23 2026, replayed at 16x from the signed ledger and the fifteen child transcripts: fifteen copies in 36.6 seconds, the next three attempts refused at the ceiling, then five more refused when the copies tried to launch their own, 29,997,622 tokens at $78.21 list price (not a bill). The opening card states the result; every time on screen is recorded time; the token fill is drawn between each copy's recorded start and end. The animated card on /try is a prepared session; this clip is not.

The free starting point

See where your session's tokens went.

Burn 0.3.1 explains the current session with npx @agentguard-run/burn why. Replay looks back at your local history and shows where the runaway-session circuit breaker would have stopped.

One of our machines · 2026-09-23 (America/Panama) · 393 sessions

npx @agentguard-run/burn replay

27.94B

Observed tokens

1.03B

Tokens after the first STOP point

3.7%

Share of recorded tokens

2 STOP · 7 WARN · 384 clean · 764 spawns

A local replay snapshot, not a billing claim or a share of the bill. The post-STOP tail is an upper bound. Replay's legacy counters can differ from the response-reconciled why report. Read the generated totals.

Explore Burn why and pace

Your policy, at the tool boundary.

Enforce on one machine

Enforce local rules, record signed decisions and use Burn for free. No key or account is needed. Shadow is a fallback state or an explicit local policy choice.

Compare free and paid

Apply the session rules

Free enforces tool allowlists, ethical-wall denies and configured budgets locally. Solo adds three machines, the dashboard and receipts export. Team adds org policy and seat management.

Configure the plugin

Know the boundary.

The plugin covers local tools sent through the Codex and ChatGPT Work hook path, including plugin MCP tools. It does not cover web ChatGPT or hosted tools such as WebSearch. Hooks are fail-open on errors and timeouts.

Read coverage and limits · Enterprise installation

English · Español · Português (Brasil)