sudo-ai-git

Hol Guard AWS Preflight — DevOps skill for Claude Code

DevOps community

Fail-closed pre-execution HOL Guard skill for agents running aws CLI commands.

How to install Hol Guard AWS Preflight

This entry records only its repository, not the path inside it, so there is no exact command to give. Open sudo-ai-git/hol-guard-aws-preflight and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Hol Guard AWS Preflight does

Fail-closed pre-execution HOL Guard skill for agents running aws CLI commands. Implements the aws/agent-toolkit-for-aws#270 feature request (build/demo reference). Uses HOL Guard's real install/run/doctor CLI contract; routes aws through hol-guard run ; only explicitly-safe verdict runs.

Alternatives in DevOps

Composio Split — Manage Split feature flags and experiments 44.6k ★
Cccc — Coordinate your coding agents like a group chat — read receipts, delivery tracking, and remote ops from your p 1.1k ★
Cloudflare Skill — Comprehensive Cloudflare platform reference docs for AI/LLM consumption 709 ★

README

HOL Guard AWS Preflight Skill

A fail-closed, pre-execution safety skill: coding agents that run `aws` CLI commands route them through [HOL Guard](https://github.com/hashgraph-online/hol-guard)'s local runtime layer. Only an **explicitly-safe** Guard verdict permits execution; every review/risky/unknown/error/blocked state is a stop condition.

This is a working implementation of the feature requested in [aws/agent-toolkit-for-aws#270](https://github.com/aws/agent-toolkit-for-aws/issues/270) ("optional HOL Guard skill for pre-execution AWS CLI safety"), offered as a public build/reference since that repo is not accepting external code contributions at this time.

Why it exists

Coding agents (Claude Code, Codex, Cursor, Kiro, Copilot CLI, …) build AWS workloads **and** can execute `aws` CLI commands from their shell tool. AWS-native controls (IAM, CloudTrail, the managed MCP Server) are the first line; this skill adds a **local, pre-execution** guard on command-bearing agent actions so a destructive `aws` command is stopped/reviewed before the harness runs it.

It does **not** replace IAM and does **not** intercept the managed AWS MCP Server.

Fail-closed decision

Guard verdict	Agent action
allow / safe / explicitly-safe	proceed with the exact `aws` command
review / approve-pending	STOP — do not run; surface approval path
risky / blocked / deny	STOP — do not run; report guard reason
unknown / error / unavailable	STOP — do not run; run `hol-guard doctor`

Contract

Setup: pipx install hol-guard then hol-guard init (auto-detects agents, prints plan, one approved change at a time).
Install for the real harness: hol-guard install (e.g. hol-guard install codex) — use Guard's own flow, never guess adapter names.
Health gate (required): hol-guard doctor must pass before any protected command.
Execute through the guard: `hol-guard run --