Skip to content

Add aws-routing skill - #81

Open
midakin wants to merge 2 commits into
aws:mainfrom
midakin:feature/aws-routing
Open

Add aws-routing skill#81
midakin wants to merge 2 commits into
aws:mainfrom
midakin:feature/aws-routing

Conversation

@midakin

@midakin midakin commented Sep 10, 2026

Copy link
Copy Markdown

Add aws-routing skill

Summary

Adds aws-routing, a read-only AWS routing subject-matter-expert skill for path selection, BGP
traffic engineering, and route evaluation across AWS networking services (Cloud WAN, Direct Connect,
Transit Gateway, VPC, VPN). It traces the end-to-end traffic path, applies the documented
route-evaluation order for each construct, flags non-deterministic path selection, and recommends
deterministic alternatives — grounded in public AWS documentation and field-proven patterns. It is
strictly advisory: analysis, recommendations, and read-only validation commands
(describe-* / get-* / list-*) only, never mutating actions.

What gap it fills

DevOps Agent can reason about individual routing facts but, in baseline testing, missed several
routing-specific determinants: the Cloud WAN CNE "deterministically random" final tiebreaker, the
DX-location-to-region association nuance, native Cloud WAN DX-attachment filtering behavior, and
TGW-to-VPN scaling paths. This skill encodes that route-evaluation logic so answers are deterministic
and correctly scoped.

What's included

skills/aws-routing/
├── SKILL.md            # activation metadata + route-evaluation instructions
├── README.md           # usage + non-production disclaimer
├── CHANGELOG.md        # version history
├── references/         # Cloud WAN/DX, DX, TGW, VPN+DX, strategic-questions patterns
└── evals/              # evals.json, eval_queries.json

Testing

  • Manually validated in AWS DevOps Agent (skill activates from natural-language routing prompts;
    reasoning trace confirms invocation). Baseline vs with-skill comparison captured.
  • Agent Skill Eval (aws-samples/sample-agent-skill-eval): Audit A (98/100, 0 critical / 0 warning),
    Functional A (0.972), Trigger 100% (14/14, standalone run). Run artifacts are regenerated locally;
    the submission ships the spec-required evals.json + eval_queries.json.

Checklist

  • Follows the Agent Skills specification
  • SKILL.md has version and author in metadata
  • name frontmatter matches directory name (aws-routing)
  • description states what the skill does and when to activate
  • README.md includes the non-production disclaimer
  • CHANGELOG.md present
  • Tested in DevOps Agent — activates consistently
  • Agent Skill Eval passing
  • No internal Amazon tool names, wiki links, or SIM tickets in submitted content
  • No customer data, account IDs, or credentials
  • Rebased on latest upstream/main

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant