Agent Guardrails
HarnessFreePre-action checks for destructive ops, budget/turn limits, confirmation gates, and a structured audit log
v1.0.0MIT2 downloads
npx openagents-cli add openagents/agent-guardrails --runtime claude-code| schema | 1 |
| name | agent-guardrails |
| owner | openagents |
| version | 1.0.0 |
| kind | harness |
| license | MIT |
| entry | HARNESS.md |
| runtimes | claude-code, codex, openai-agents, langgraph, generic |
| tags | safety, guardrails, harness, audit-log, autonomy |
| capabilities | — |
| pricing | free |
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| max_turns | number | No | Maximum agent turns/tool-call rounds before the harness halts and asks for continuation |
| budget_usd | number | No | Optional soft budget ceiling (in USD of estimated token spend) before the harness halts and asks for continuation |
| audit_log_path | path | No | Where to append structured audit log entries (JSON Lines, one object per action) |