product updateAmazon Web Services

AWS Ships Six Agent Skills to Automate Amazon Bedrock's Automated Reasoning Policy Lifecycle

TL;DR

AWS published a suite of six Agent Skills that automate the full lifecycle of Amazon Bedrock Automated Reasoning policies—from rule extraction to deployment—directly from coding agents like Claude Code, Cursor, Kiro, and Codex. The skills wrap Bedrock's formal-logic verification APIs in structured workflows built on Anthropic's open Agent Skills format.

3 min read
0

What happened

AWS released a suite of six Agent Skills that let developers run the entire Amazon Bedrock Automated Reasoning policy lifecycle—authoring, testing, debugging, deployment, and validation—directly from a coding agent instead of the Bedrock console. The skills are open-source and available now in the amazon-bedrock-samples GitHub repository under responsible_ai/automated-reasoning-checks-skills.

Amazon Bedrock Automated Reasoning checks validate AI outputs against formal logic rather than statistical sampling, giving what AWS describes as mathematical certainty that a response complies with a defined policy. The system works in two stages: foundation models first translate a question and answer into formal logic, mapping natural language to policy variables; then an SMT (Satisfiability Modulo Theories) solver checks that logic against the rules and returns a verdict with supporting or contradicting evidence.

The six skills

Each skill corresponds to one stage of the policy lifecycle:

  • Builder — creates a policy from a source document and extracts rules, variables, and custom types
  • Reviewer — parses quality and fidelity reports to flag conflicting rules, unused variables, and bare assertions
  • Tester — generates test scenarios and runs question-and-answer checks against the policy
  • Debugger — diagnoses failing verdicts, working from the premise that most errors stem from translation issues rather than the rules themselves
  • Deployer — snapshots a numbered policy version and attaches it to a guardrail
  • Validator — calls the ApplyGuardrail API and can run an automated rewrite loop, feeding contradicting rules back to the model until an answer passes

Each skill follows the standard Agent Skills layout: a SKILL.md instruction file, a references/ folder with API and rule-syntax detail loaded on demand, and a scripts/ folder of standalone Python scripts that support --help and --dry-run flags for inspecting requests before they hit Bedrock. A shared library handles client creation, build-workflow polling, and finding parsing across all six skills.

Format and compatibility

Agent Skills is an open format created by Anthropic for packaging domain-specific instructions and workflows into coding agents. Because the format is open, AWS's skills install into any compatible agent, including Claude Code, Kiro, Cursor, and Codex. Installation runs either through Claude Code's plugin marketplace (/plugin marketplace add) or via npx skills add pointed at the GitHub subdirectory for other agents.

In a worked example using a short HR parental-leave policy document, AWS reports the builder skill extracted six rules, four variables, and one custom type from the source text in a single build run, illustrating the intended end-to-end flow from plain-language policy document to a deployed, testable guardrail.

What this means

This is not a new model or a pricing change—it's tooling that lowers the friction of an existing Bedrock feature. Automated Reasoning checks have a real learning curve: authoring correct SMT-LIB rules and tuning variable descriptions so translation matches user intent is detail-heavy work with a build-test-refine loop of its own. By packaging that judgment into six discrete skills with runnable scripts, AWS is betting that coding agents are now reliable enough to own repetitive, rule-bound verification workflows rather than just code generation.

The more notable signal is AWS building on Anthropic's Agent Skills format rather than a proprietary AWS mechanism. That choice makes the skills portable across Claude Code, Cursor, Kiro, and Codex out of the box—a tacit acknowledgment that Agent Skills is becoming a de facto standard for distributing agent workflows, independent of which lab built the underlying coding agent. For teams already using Automated Reasoning checks for compliance-sensitive domains like HR, lending, or insurance eligibility, this shifts policy management from a console-driven, expert-collaboration process toward a code-reviewable, version-controlled one—likely the more durable requirement for regulated deployments over time.

Related Articles

product update

Kimi Code CLI v2.7.0 adds native Amazon Bedrock provider with standard AWS credentials

Kimi Code CLI v2.7.0, Moonshot AI's open source coding agent, adds a native bedrock provider. It resolves the standard AWS credential chain, including named profiles and SSO, so users no longer need a separate Amazon Bedrock API key. AWS demonstrates the setup with Kimi K3.1 Pro Highspeed.

product update

Amazon Polly voice cloning is generally available as a gated release in seven languages

Amazon Polly voice cloning is now generally available as a gated release. It builds a custom voice on the Polly generative engine from a few minutes of one consenting speaker's audio, in seven languages. Building a voice is free, and usage is billed pay-as-you-go.

product update

Zed v1.29.1 adds Claude Fable 5.2 for Anthropic BYOK users and a Git Panel branch mode

Zed v1.29.1 adds support for Claude Fable 5.2 for Anthropic BYOK users, plus Claude Mythos 5.2 for organizations with access. The release also adds a Git Panel branch mode, a bookmark picker, configurable Git date formats and a setting to turn off automatic agent thread titles.

product update

Cline v4.1.27 fixes reasoning-effort errors that caused GPT-6 and gpt-5.6 tool-call rejections

Cline v4.1.27 fixes several reasoning-effort bugs that caused GPT-6 and gpt-5.6 endpoints to reject tool calls. It also refreshes the model catalog with 94 new providers and changes default models for 67 others.

Comments

Loading...