release-sandbox-tools

Builds, validates, and publishes inspect-sandbox-tools binaries to S3 for PR landing.

2.7k|685|Updated Nov 14, 2023
One-click install
npx skills add https://github.com/UKGovernmentBEIS/inspect_ai --skill release-sandbox-tools
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: release-sandbox-tools
Source: https://github.com/UKGovernmentBEIS/inspect_ai/tree/main/.claude/skills/release-sandbox-tools
Command: npx skills add https://github.com/UKGovernmentBEIS/inspect_ai --skill release-sandbox-tools

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

Landing a PR that changes code under src/inspect_sandbox_tools/ fails CI because the slow-tool-tests-release check requires newly versioned injectable binaries to be built, validated, and published to S3 — a step contributors cannot perform without maintainer AWS credentials.

Core Features & Use Cases

  • Interactive release wizard: Runs scripts/release-sandbox-tools.sh to build four artifacts (amd64/arm64 × glibc/musl), validate them across Linux distros, upload to S3, and commit pinned SHA256 digests back to the PR branch.
  • Agent-safe auto mode: The --auto flag resolves confirmations to safe defaults and aborts with instructions when human action (Docker startup, aws sso login) is needed.
  • Resumable execution: Re-runs skip completed build and validation stages when a recorded fingerprint matches the current source, making dry-run-then-real-run workflows cheap.
  • Use Case: A maintainer needs to merge an approved contributor PR whose only failing check is slow-tool-tests-release; they run the wizard from a worktree of the PR branch, then watch CI go green after the digest commit is pushed.

Quick Start

Run the release-sandbox-tools wizard in auto mode from a checkout of the approved PR branch to build, validate, and publish the sandbox tools binaries.

Frequently Asked Questions about release-sandbox-tools

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I land a PR that changed inspect_sandbox_tools code?

Run scripts/release-sandbox-tools.sh --auto from a checkout of the PR's head branch once the PR is approved and slow-tool-tests-release is the only failing check. The wizard builds, validates, uploads the binaries to S3, and pushes the pinned SHA256SUMS digests.

Why does the slow-tool-tests-release CI check fail on my PR?

It fails because the bumped sandbox tools version's artifacts are not yet published to S3 with committed digests. The check fails at the fetch-and-verify step, either on the SHA256SUMS lockstep check or with a message naming the missing S3 object.

Can I run the release wizard without publishing anything?

Yes, pass --dry-run to build and validate the binaries without uploading to S3 or pushing digests. Because runs are resumable via recorded fingerprints, a later real run skips the slow stages already completed.

What credentials are needed to publish sandbox tools binaries?

The upload uses your ambient AWS session, so run aws sts get-caller-identity first and aws sso login if credentials are stale. Typically a maintainer performs this step since contributors lack S3 upload credentials.

What happens if binaries for a version already exist on S3?

Published versions are immutable and the upload script aborts if an object exists with different bytes. Never force-overwrite; instead bump the version and publish fresh artifacts built from the exact PR code.