agentic-infra-claude-skill
A Claude Code skill for designing, reviewing and linting agentic AI infrastructure, built on 21 core design patterns, a review checklist and a deterministic ADR linter.
As of October 2026.
The skill triggers when a task is about an agentic system. It picks patterns for a new design, audits an existing one against the checklist, or lints an ADR. The core patterns and checklist items cite pages of one source PDF, so a finding can be traced to its reasoning; the later community patterns cite their own sources.
AgentForge was self-assessed against the 62 core items; its review file maps each one to the design document that covers it. This site's own agent-side decision records are written on the skill's lintable ADR template (self-reported).
Details
Results
Covers 21 core agentic design patterns, from prompt chaining and routing to guardrails, evaluation and exploration, each citing pages of its source PDF; ten community patterns were added later.
you can check this: agentic-infra-claude-skill
Reviews an ADR or design document against an architecture checklist of 62 core items tied to the source PDF, 76 with the sections added later.
you can check this: agentic-infra-claude-skill
A deterministic linter checks an ADR markdown file against 17 rules, prints JSON and exits 1 on CRITICAL findings.
you can check this: agentic-infra-claude-skill
How is an architecture decision checked mechanically?The AgentForge design was self-assessed against the checklist's 62 core items and met all of them.
you can check this: AgentForge